Visualization Using Rational Morphology
Authors: Robert Kogan, Dr. Sos Agaian, Dr. Karen Panetta Lentz
Phone: +1-617-6273217
E-mail:
In this paper, Rational Morphological Transforms are investigated and employed for detecting and visualizing objects. Previous efforts in this area have used magnitude reduction techniques applied to the entire image.
It is shown that by attenuating the magnitude of an image based on the morphology of the image, better detection and visualization can be acheived. Further research is conducted on combining a bank of morphological filters in parallel, and analysis is done on the performance benefits versus the increase in complexity.
Finally, several methods are implemented to increase the speed of the morphological filters for real-time analysis of images.
Key words: rational morphological filters, magnitude reduction, detection, visualization
Adaptive compression of image data
Authors: S. Hludov, C. Schr?ter, Chr. Meinel
Phone: +49-651-97551
Fax: +49-651-97551-12
E-mail:
Superhigh quality colour image TV system TELAN
Author: A. Antonov, Telan Ltd
Phone: +380 44-4443587
Fax: +380 44-4443595
The disadvantage of the systems of television of high sharpness is unsettledness of the problem of noise reducing. The disadvantage of the systems of digital television is unsettledness of the problem of exceeding of sharpness of reproduced image. Their common disadvantages are the bad compatibility with the existing standard television systems, high complication and cost. That's why the systems of digital television of high sharpness were proposed.
The perspective alternative to such systems is the system of television of high quality TELAN, free from all the above mentioned disadvantages. This system when accepting signals from any existing stations of the standard television provides sharpness of image up to level, limited by the solution ability of cathode ray tube, provides additional reducing of the high frequency noise to 20 dB, has practically the same cost as the presently produced television techniques. Such results are achieved due to the usage of psycho-physiological redundancy of the starting analog videosignals.
Key words: videosignal, noise, restoration, recovery, redundancy
The Practice of Multispectral Image Acquisition
Author: Friedhelm König
Phone: +49/241/80-7718
Fax: +49/241/8888-198
E-mail:
One of the major problems of image acquisition by cameras or scanners is the metamerism of the camera, which differs from the metamerism of the human eye. This is caused by the deviation of the sensor response from the responses of the human eye [1,2]. A solution to this problem is the capture of the entire spectrum of a color by a multispectral camera rather than just measuring a red, a green and a blue component. In this paper a practical realization of a multispectral camera is described.
Actual implementations of multispectral image acquisition devices are proposed by Keusen, Burns or Tsumura[3,4,5]. This paper focuses on essential design parameters of a multispectral camera, consisting of a CCD-array, a spectral filter set, a defined light source and optics for the capture of reflective images.
Essential points are:
Position of the filters: There are two locations in the optical system, where the filters can be positioned, between the light source and the object to be measured, or between the object and the camera. The advantages and disadvantages of both positions are discussed. In the current realization the second configuration is chosen, and our experiences with this configuration will be described in detail.
Number and bandwidth of the filters and interpolation algorithm: These parameters interact with each other. The choice of the number of filters and their bandwidth on the one hand forces the choice of the interpolation algorithm on the other hand.
In our configuration average color errors below 2 CIELAB units are achieved (Number of filters: 16, bandwidth: 10nm, interpolation algorithm: smoothing inverse [6]) .
The discussion of design parameters is based on an experimental multispectral camera designed at the the Technical Electronics Institute. Several typical sources of errors will be discussed (errors caused by imperfection of the light source, the shutter, and the filters).
Finally, experimental results will be given to demonstrate the capabilities of the concept.
References:
[1] Preserving original image content in high-resolution digital archives,
Mark H. McCormick-Goodhart, Proc. IS&T's 50th Annual Conference, 1997,
pp. 584-587
[2] Reconstruction of natural spectra from a color sensor using nonlinear
estimation methods, Proc. IS&T's 50th Annual Conference, 1997, pp. 454-457
[3] Multispectral color system with an encoding format compatible to the
conventional tristimulus model, Thomas Keusen and Werner Praefcke, Proc.
IS&T/SID 1995 Color Imaging Conference, 1995, pp. 112-114
[4] Analysis of image noise in multispectral color acquisition, Peter D.,
Burns, Dissertation at the Rochester Institute of Technology, 1997, pp.
77-91
[5] Optimum spectral transmittance of color filters on single chip CCD for
electronic endoscope, Norimichi Tsumura et. al., Proc. IS&T's 50th Annual
Conference, 1997, pp. 373-376
[6] Digital color image restoration, Clanton E. Mancill, Dissertation at
the University of Southern California, 1975, pp. 56-70
Key Words: Multispectral Imaging, Error Sources, Practical Implementation
Biography: Friedhelm König started his study of electrical engineering at the Aachen University of Technology in 1990. He achieved his diploma degree in 1996 and graduated summa cum laude. His diploma thesis dealt with spectral estimation algorithms. Now he is a scientific assistant at the Technical Electronics Institute of the Aachen University of Technology. He is working on multispectral techniques and related subjects.
Economical subject-selective metrics of perceptible color image quality
Author: V. Gavrik
Phone: +49-220335641
Acousto-optical tunable filter spectral resolving power measurement principles
Author: Vasily Popkov
Phone: +7-812-2518839
E-mail:
Acousto-optical tunable filters (AOTF) are used in the cases when we need to select a set of certain colours from the total multicolour image or to perform spectral analysis in real time. The most important characteristic of AOTF is their spectral resolving power which describes the wavelength difference between two distingushed spectral lines.
Each researcher performing investigation of any AOTF notes the value of its spectral resolving power but the resolving power criterion is practically never minded. This is the reason of the fact that the results of the spectral resolving power measurements of the same device made in different laboratories, may differ from each other.
The spectral resolving power criterion based on the information approach to AOTF operation has been developed, and the calculations similar to those made for spatial resolving power of acousto-optical devices [1] have been carried out.
Taking into consideration the developed criterion, the spectral resolving power measurements setup has been constructed. The measurements have been carried out using specific means for the noise level determination, and the accompanying calculations of the spectral line distinguishing probability have been made on the basis of the experimental data.
If the problem to be solved by AOTF requires the neighbour spectral lines intensities comparison, the gray scale levels must be transmitted by the considered device. It has been shown that the resolving power depends strongly on the amount of gray scale levels as well as on probability of their distinguishing.
The experimental results demonstrate that information approach may be successfully applied to the AOTF spectral resolving power metrology. It has been shown that it is strictly connected with amount of information transmitted and processed by AOTF.
References
[1] V.N.Sokolov, B.S.Gurevich, S.V.Andreyev, P.A.Burov, and V.A.Markov,
"Acousto-optical devices metrology: information approach", European
Optical Society Topical Meetings Digest Series, 1997, vol. 15, pp. 55-58.
Acousto-optical colour separation in the process of transmission of colour images drawn over colour background images
Authors: V.I.Popkov, N.I.Migay
Phone: +7-812-2518839
E-mail:
Among the problems connected with the colour image transmission, the following one has a specific importance. Let us represent a colour object (for example, a map) which is located at the transmitting side, and the same colour object takes place at the receiving side. Then a person at the transmitting side makes some colour signs over this object (arrows etc.). The problem is to transmit these signs to the same places of the object in the receiving side. In order to realize it, the colour separati on must be performed before the information transmission from one side to another.
The signal containing data on the overdrawn signs must be introduced into the electronic transmission network (radio, electronic mail, or something else). The most important factor creating obstacles to this signal transmission is the presence of undesired data on the initial object colors distribution. We suppose to apply an acousto-optical tuning filter (AOTF) for the colour separation.
The proposed AOTF has been prepared on the basis of Bragg cell which included active medium made with tellurium dioxide single crystal and piezoelectric transducer based on LiNbO2 plate. The TeO2 crystal was cut in such way that its axis was deflected by 6.50 from [110] crystallographic axis. The transducer excited slow shear acoustic wave in the crystal. If the crystal is illuminated by the collimated polychromatic light, it selects the light with certain wavelength which corresponds to the frequ ency of the acoustic wave excited in the crystal.
Experimental investigations may be carried out under the condition if the overdrawn signs were made with the colour pencil which colour is close to monochromatic. If this condition is met, the AOTF is conjugated with this colour (the certain acoustic frequency is excited).
The experimental study of the colour selection process have been carried out using the experimental setup which included initial colour object with overdrawn colour signs, optical circuit which provided information transmission to the AOTF, Bragg cell, corresponding electronic circuits providing RF input signal, output optics, and photodetector array.
The results of experiments have shown that data on the overdrawn signs were successfully transmitted to the output photodetector. The analysis of output signal have demonstrated that the influence of initial object colour distribution data (which represents an obstacle relatively to the considered signal) is negligible.
Key words: acousto-optical tunable filter, colour images transmission, separation of colours
LED-illuminated CCD slide scanner with a new concept for color processing
Authors: B. Hill, M-C. Kim
Phone: +49-241-807700
Fax: +49-241-8888198
A Fast Edge-Preserving Image Interpolation System with an Arbitrary Aspect Ratio
Authors: Hwang-Cheng Chiang and Wei-Feng Hsu
Phone: 886-3-591-3063
Fax: 886-3-582-9781
E-mail:
Conventional interpolation techniques, including pixel replication, bilinear interpolation and spline based methods have been popularly used in commercial applications. It is often desired that the interpolation process is able to accurately and fastly increase the spatial resolution of an image at an arbitrary aspect ratio and with sharp edges. These do not generally happen to conventional algorithms, which consume a great effort to interpolate an image at an arbitrary ratio and tend to blur edges or introduce blocking artifacts. In this paper, we propose an interpolation scheme which is more general than the existing ones. The method is based on the fact that the Fourier transform (FT) at an arbitrary frequency can be expressed as a weighted sum of its Discrete Fourier transform (DFT) coefficients. These weights can be suitably approximated so that the FT is very close to the sum of (1) a few dominant terms of the sum of the DFT coefficients, and (2) the DFT of a new sequence obtained by multiplying the original sequence with a sawtooth function. If we take the inverse discrete Fourier transform (IDFT) of an image; then by the algorithm described above, the spatial sample at an arbitrary location can be fastly computed by the fast Fourier transform (FFT) algorithm. In addition to retaining the computational efficiency of the FFT algorithm, experimental results have revealed that the proposed method preserves the sharp edges of the original image.
Key words: Interpolation, digital zoom, DFT, FFT
Biography: Dr. Chiang received the Ph. D degree in electrical engineering from Tatung Institute of Technology, Taipei, Taiwan in 1996. Since October 1996 he has been with the Image Technology Department, Opto-Electronics and System Laboratories/Industrial Technology Research Institute. His current research interests include digital image processing, multirate signal processing and digital filter design.
Image Analysis in Multicolour Printing
Authors: Marie Kaplanova, Assoc. Prof. Ph.D. and Miroslav Fribert,
Ing. Ph.D.
Phone: +420-40-41 011
Fax: +420-40-42 209
E-mail:
The poster will introduce some works from the region of image analysis of multicolor printed samples. These samples have been scanned by the CCD camera from printing sheets, which have been printed by usually used printing technologies (ofset, gravure, flexografy). Certain image analysis methods such as statistical processing of colour values, image smoothing, image contrast enhacement, segmentation of halftone dot regions of process inks and segmentation of process inks overprints were used in the processing of multicolour print samples.
The results of these works can used to evaluate printing parameters e.g. halftone values, denzities in the screen dot regions and trapping in overprint regions. The evaluated values of printing parameters can be then used to control the quality of printing process.
Key words: CCD camera, image analysis, colour printing, printing parameters, print quality.
Biography: Marie Kaplanova;, Associate Professor, MSc., Ph.D., born 1941 in Prague, 1962 graduated from Charles University Prague in Physics, 1977 Ph.D. in experimental physics, 1990 Associate Professor in Technology of Fibrous Materials, from 1992 the Head of Department of Graphic Arts and Photophysics. The aim of the science-reseach activity: photoacoustics, photocalorimetry, fluorescence spectroscopy, SALS. Member of Science council of FCHT University Pardubice, member of European Society for Photobiology, The Society for Imaging Science and Technology, representative of the department of graphic arts and photophysics in International Association of Research Institutes for the Graphic ArtsIndustry. Author of more than 80 scientific papers (cited in about 84 papers according SCI), author of 4 textbooks.
Better lossless compression for palette colour images
Authors: Vaclav Hlavac, Jaroslav Fojtik
Phone: +420 2 24357465
Fax: +420 2 24357385, hlavac@vision.felk.cvut.cz
Background
Our contribution relates to lossless image compression. Specifically, we
tackle pseudo colour images (images with palette).
Materials and Methods
We propose a new greyscale transformation that allows better compression
ratio from palette colour images. The use of the compression algorithm that
treats colour images as bit planes is assumed. This image transformation
was tested with the compression algorithm: Hlavac, V., Fojtik J.: Adaptive
non-linear predictor for lossless image compression, Proceedings of the
conference Computer Analysis of Images and Patterns'97, Kiel, Germany, Springer-Verlag,
LNCS Vol. 1296, September 1997, pp. 279--288. but should work with others
as well.
Let us assume the palette image palette(f(x,y)). The algorithm that quantized the original true colour image caused discontinuities in index function f(x,y). This effect decreased the compression ratio. Our idea is to re-establishes the continuity of the index function f(x,y). We change both the index function (f'(x,y)) and the palette' in the way that the resulting value of the image function for all x,y remains the same palette(f(x,y)) = palette'(f'(x,y)). Each multi-level image can be treated as stacked bit planes. Our method minimizes the number of value changes in the current bit plane by changing bit planes below it. We start from most significant bit plane and then the algorithm proceeds to bit planes that are below it. First of all, the table of neighbourhood relations of all colors (indices) is computed and stored in a table (triangular due to symmetricity of the relation). The set of all indices is divided into 2 groups. The quality of each index can be calculated as number of indices in the own group minus number of indices the other group. Global optimization criterion is sum of qualities of all indices in both groups. The worst index in each group is found. The worst indices between groups are swapped. This is repeated until the optimization criterion decreases.
Results
We tested the method on real images and the gain in bit saving is about
10%. We believe that the method can be easily incorporated into standard
lossless compression methods.
Biography: Vaclav Hlavac is the Associate Professor and the head of the Center for Machine Perception of the Czech Technical University in Prague. Born 1956, PhD 1987. His research interests cover image analysis, 3D vision and industrial applications of computer vision. Jaroslav Fojtik is a PhD student.
Colour image comparison and visual perception : a process for descriptor validation
Authors: Eric Favier, Eric Dinet and Alain Tremeau
Phone: +33 477 92 30 30
Fax: 33 477 92 30 39
E-mail:
Key words : Colour image comparison, Visual perception, Colour image quality
Gray scale image processing provides very satisfying and original answers to various problems in many fields like quality control, medicine, geology, security, etc. But, in some problems, a gray scale analysis is unfortunately not sufficient : colour is necessarily required. For example : to judge the quality of colour printings in textile industry, to control colour drifts in printing works or to evaluate degradations introduced when compressing colour images. Generally in these kinds of applications it is necessary to be able to quantify the quality of an image in accordance with a reference one, the quality of a process in accordance with another one, the evolution of the quality of images obtained by a particular algorithm according to the settings.
Such quantifications must provide results correlated with human vision since colour image quality is generally evaluated by human observers. Nowadays no objective descriptors and only few relevant criterions are available for colour images. Moreover these criterions are, for most of them, global descriptors which have not the ability to highlight local degradations. However these local degradations are crucial in the response of human observers.
Even if in recent research a major emphasis has been given to a deeper analysis of human visual system, experimental or theoretical studies concerning the comparison of image quality measures with subjective visual perception are not in general use. Therefore, in a first time, we developed objective local descriptors for the evaluation of quality of colour images. Then we established a process to validate the accuracy and the correlation of these descriptors with human visual perception.
The goal of the proposed report is to expose how we determine during an experimental process :
We propose to describe the whole experimental process used to study these correlations. We will also expose results obtained through the experimental process and we will illustrate our approach with compression techniques.
Biography: Eric FAVIER received the PhD degree in image processing in 1994 from the University Jean Monnet of Saint-Etienne. His thesis work dealt with the use of multiresolution tools in computer vision, with examples in discrete contour analysis. He is currently a research fellow in the Vision Engineering group managed by Prof. Bernard Laget and is involved in aspects related to colour and texture analysis.
Color halftoning by indexing the visual-optimized dot profiles
Author: H.W. Park
Phone: +82-2-9583318
Fax: +82-2-9602103
E-mail:
Retrieving similar colour images
Authors: Dr. R. Schettini
Phone: +39-2-70643288
Fax: +39-2-70643292
E-mail:
There is a great demand for efficient tools that can, on the basis of the pictorial content, organise large quantities of images and rapidly retrieve those of interest. With that goal in mind we present a method for indexing complex colour images. The basic idea is to exploit image data decomposition and compression based on the standard Haar multiresolution wavelet transform to describing image content. In this way we are able to effectively eliminate data redundancy and concisely represent the salient features of the image in image signatures of predefined lengths. In the retrieval phase image signatures are compared using a similarity measure that the system has "learned" from users. Experimental results confirm the feasability of our approach, which outperforms more standard procedures, in retrieval accuracy and at lower computational costs.
Characterization of color texture using image analysis - Color and texture based sorting of tiles.
Authors: BOURADA Y.,LAFON D.,ETERRADOSSI O.
E-mail:
The aim of the present work is the analysis of the color texture of tiles (a "4?4 cm¦" surface) in order to achieve an automatic sorting based upon their appearance.
The tiles color texture can be seen as a red background over which dark regions are superposed. These regions contain occasionnally little black grains. For the sake of the study, two sets of tiles are selected :
A first sorting of tiles is obtained by means of a spectrocolorimeter. It doesn't take texture into account.
A 3CCD color detector is used for digital acquisitions of images. The acquisition system is colorimetrically calibrated. It allows to obtain color images coding in the CIEL*a*b* system. The accurate measurement of colors provides, when quantifying a texture, parameters directly linked to colorimetric features. The obtained values are directly correlated to the human perception.
The selected texture characteristics are :
The different obtained parameters are used for the definition of a metric which is in a good agreement with an eye sorting.
Color characterization by camera is less accurate than the one obtained by a spectrocolorimeter (10% of maximum relative error in color measurement). Nevertheless, a color texture based characterization leads to make more efficient aspect based sorting operations.
Characterization of color texture - ciel*a*b* calibration of ccd device.
Authors: Laflaquiere Ph., Lafon D., Eterradossi O., Slangen
P.
E-mail:
A lot of materials used in the construction industry are colored materials showing strong color texture, which may give the product its commercial value. Only the eye is currently used to qualitatively control appearance variability of such products. We have planned to perform automated appearance sorting using a tri-CCD color video camera connected to a true color image digital frame grabber.
The aim of the present work was the refinement of a calibration process allowing this tool to deliver for each pixel the same information as a colorimeter, in order to make quantitative color texture analysis of the observed materials. To perform the planned calculations, RcamGcamBcam values needed to be converted into values belonging to a uniform colour space.
Analysis of the acquisition tool first allowed characterization of the card and camera behaviour. Linearity of R, G and B gains has been tested using monochromatic laser lights (476.5 nm - 514.2 nm - 632.8nm) of varying intensities. Identification of the different noise sources has been performed. Spatially varying noise influence upon the precision of color value acquisition was evaluated. Fixed pattern noise was corrected using a mask.
Color calibration was performed using measurements with a spectroradiometer . The values used in this step are tristimulus values (X,Y,Z). The results are shown to be in good agreement with expected results.
Taking the illuminant characteristics into account allows transformation to the CIE system (either L*a*b* or LCH) with a maximum relative error of 10%.
After such a calibration, the acquisition system becomes a measurement tool only needing to be calibrated again if any of the card or camera characteristics is changed (gain, offset, white balance).
Quality Assessment of Digitised Aerial Photographs
Author: O. Koelbl
Fax: +41 21 693 57 20
E-mail:
Manufacturer of photogrammetric instruments have developed specific scanners for aerial photographs, in parallel to printing industry. Main objective of this specific scanners is to guarantee a high geometric precision of ¤0.001 to 0.002 mm for a stan-dard format of original film documents of 23 x 23 cm and to scope with the high im-age resolution of the original images. It is understood that also a high dynamic range of the scanner, low image noise and a good colour reproduction would be desirable. Contrary to printing industry, the fidelity of tone reproduction is of minor impor-tance, much more important is an appropriate image enhancement, as for colour and for grey tones in order to assure a good object recognition. Effects of image distur-bance like haze or poor image contrast should be corrected.
Within a study of OEEPE (European Organisation for Experimental Photogrammet-ric Research) the most important photogrammetric scanners used in practice have been tested. Standard procedures are in development to analyse the dynamic range of the scanners, the image noise, the image sharpness and the colour fidelity. Practi-cal all photogrammetric scanners are based on CCD technology. Consequently the dynamic density range is limited to about 1.5 D, the image noise remains within this range under 0.03 D. As for image sharpness the digitised images can not yet scope with photographic enlargements and the modulation transfer function of the scanner does not match with the MTF of the original images. Consequently the digitised im-ages can not yet replace the plotting techniques on high quality photogrammetric equipment using the original photographs.
The article presents the techniques applied for the testing of the scanners concerning the determination of the MTF of the scanners, the image noise, the dynamic range and the colour fidelity and gives typical results for various scanners. Furthermore the article shows approaches for image improvement as for colour reproduction includ-ing the incorporation of false colour photographs and their conversion into colour photographs. The scanners tested are manufactured by Intergraph, Zeiss, Agfa, He-lava, Wehrli and others.
Automatic Identification and Virtual Reality Modeling of the Lungs.
Authors: Jamshid Dehmeshki, Fraser Hatfield, Farhang Daemi.
Phone: +44 115 956 8995
Fax: +44 115 956 8991
E-mail:
Background: This paper focuses on the development of a virtual reality model of the lungs, which can be visualised and interacted with in real-time. Method For this work, a single breathhold, Spiral CT sequence was used. For each transaxial slice in the scan an array size of 512 by 512 pixels was obtained with an in-plane resolution of 1 mm. An novel technique has been developed which integrates textural characteristics and spatial information for the automatic identification of the lungs. This 3-D segmentation techniques consists of two stages. In the first stage we impose a stochastic model to a given image and the segmentation problem is then approached as a statistical estimation problem. By this we obtain the outline of both lobes of the lung. In the second stage, we apply the 3-D segmentation algorithm to the previously segmented lobes, leading to an over-segmentation of the each lobe, which can be used to detect the internal structure of the lungs. From this, the lung volume was estimated.
Also, a surface rendered fly-through model of the lungs was created. Next, endoscopic images were texture mapped onto the surface. Finally, the pulsing motion observed within the lungs was simulated.
Conclusion: We have developed an automatic method to extract the lungs from CT images. Using this technique, we have been able to estimate the lung volume accurately and create a 3-D model which can be visualised and interacted with in real-time.
Biography: Jamshid Dehmeshki received his PhD from Nottingham University in 1997, where he is currently a research fellow at CIMI. Dr Dehmeshki has participated in a variety of projects in the field of image processing. His current research is in texture analysis, Neural Networks, 3-D statistical segmentation, Fuzzy Logic, Medical image processing and real-time image processing.
Automatic Delineation and 3-D Visualisation of the Human Ventricular System Using Probabilistic Neural Networks
Authors: Dr Fraser Niles Hatfield J.Dehmeshki, M.F.Daemi, M.Vloeberghs
Phone: +44-115-956 8995
Fax: +44-115-956 8991
E-mail:
Background
Neurosurgery requires many years of training. It is now becoming common
to use virtual reality models of the complex structures within the brain.
These models can assist in training and will eventually be used in the preparation
for surgery. This will lead the way forward in the development of a fully
flexible training environment for neurosurgeons.
This paper focuses on the development of a virtual reality model of the ventricular system of the brain, which can be visualised and interacted with in real-time.
Method
For this work, a T2-weighted CISS scan was acquired using a healthy volunteer.
This sequence consisted of 177 sagittal slices of array size 256 by 256
pixels, with a 1 mm/pixel resolution. A number of regions from this data
set were used to train a Probabilistic Neural Network, which was subsequently
used to automatically segment the entire image volume. The ventricles were
then extracted from the segmented images using prior knowledge of the brain
anatomy. A three dimensional search was then performed to link together
all segments of the ventricular system. From this, a surface rendered fly-through
model of the ( that's all he sent! could you please look at the fax and
finsh the sentence? )
Structure Artefact Free Multi-Level Error Diffusion Algorithm
Authors: V. Ostromoukhov, R.D. Hersch
Phone: +41 21 693 5261
Fax: +41 21 693 6680
E-mail:
Colour ink-jet printing technology has developed at a rapid pace during the last few years. Printer manufacturers have been able to significantly decrease the volume of individual ink droplets while improving their spatial positioning. In addition, some printers use three- or multi-level inks rather than traditional bi-level inks. Nowadays, low-cost ink-jet printers reach a resolution of over 1000 dpi and provide almost photographic quality.
In early ink-jet printers, individual ink drops were relatively large and therefore clearly visible. The inherent quality of halftoning algorithms used in these printers was one of the most important factors for achieving satisfactory printing quality. Clustered-dot dithering, popular for laser and traditional offset printing, has not been used in ink-jet printers due to the relatively coarse size of the produced screen dots. In contrast, halftones produced with error-diffusion algorithms appear much finer. This is the reason why error-diffusion was used preponderantly in first-generation colour ink-jet printers.
With the advent of higher resolution (over 1000 dpi) multi-level colour ink-jet printers, the situation has changed. Dots of variable size are formed by a variable number of tiny individual ink droplets printed like small clusters.
The produced variable size dots are still invisible to the human eye. Possibly combined with multi-ink technology (e.g. low- and high-density inks, for the same colour), these new printers may be considered as multi-level ink-jet printers operating at medium resolution (say 300-600 dpi), with a relatively large number of "primary" colours. By "primary" colour we understand here different permitted combinations of dot size and/or multi ink features. The number of available primary colours, usually between 8 and 32, is not sufficient to avoid banding effects and to permit the faithful reproduction of high-quality images. Usually, a simple error diffusion algorithm is applied between consecutive primary colours. Straightforward error diffusion may produce at certain intensity levels important artefacts such as visually perturbing structures.
We first analyse the nature of the artefact mentioned above. We then propose a solution which eliminates the structure artefact problem.
Let us assume that we process separately Cyan, Magenta, Yellow and Black colour channels, halftoning each colour layer according to the value of its respective input image intensity. The whole intensity range [0..255] is subdivided into (N-1) equal intervals by the N available primary colours of the considered colour channel. A standard error-diffusion algorithm can be applied between two consecutive primary colours, in order to obtain the intermediate intensity levels. However, the structure artefacts present in bi-level standard error-diffusion algorithms (e.g. Floyd-Steinberg ED, Stucki ED etc.) will also appear when applying error-diffusion between consecutive intensity levels. The visually perturbing artefacts are situated at intensity levels close to g = 255*i/j, where i and j are small integer numbers. Typically, the intensity levels g = 255*1/2, 255*1/3, 255*2/3, etc. are the potential sources of significant structure artefacts: patches of regular structures such as checkerboards are formed. The patches' boundaries may be clearly visible, especially when amplified by the non-linear dot gain inherent to any ink-jet printing process. As mentioned, these artefacts are inherited by multi-level error-diffusion, which maps the range of intensity levels [I1..I2] between two consecutive intensities I1 and I2 onto the range [0..255].
The key idea of the proposed improved multi-level error-diffusion algorithm is to use a non-linear mapping ensuring that the corresponding two-level error-diffusion algorithm never falls into "forbidden" intensities around the dangerous values 255*i/j. Special care is needed in order to ensure smoothness of the intensity reproduction curve. The proposed method is illustrated by a set of samples, produced with our simulator which mimics the behaviour of a typical high-resolution (1440 dpi) multi-level ink-jet printer.
Polarised imaging sensor for objects detection
Authors: JM. Deler, P. Hervy
Phone: +33-1-49653581
Fax: +33-1-49653636
E-mail:
To achieve robust and efficient object recognition, particularly from Infrared real outdoor images, methods to reduce clutter and extract salient features of objects must be developed. The purpose is to find important contours corresponding to possible objects. Toward this end, a polarimetric thermal imaging technique is proposed, in order to improve the discrimination between natural and man-made object.
This paper describes the operation of the polarisation-sensitive thermal imaging approach into the MWIR spectral band. Informations on the polarisation and angle of polarisation have been gathered. A suite of polarimetric object recognition algorithms based on the fuse of the polarised and intensity informations is discussed.
Results are presented from real outdoor images for different military or commercial applications. First results in the case of Anti-Personnel mine detection are presented. It is currently considered that polarimetric thermal imaging is well adapted to detection of laid and flush mines. Then, results concerning high value target recognition are presented.
Uniform-colour scales and RGB primaries for colour-signal coding
Author: Claudio Oleari
Phone: +39-521-905214
Fax: +39-521-905223
E-mail:
Recently, chromatic-response functions with uniform scales are obtained as logarithms of proper cone-activation ratios (CAR hypothesis) and can be considered as Cartesian coordinates for equiluminant-colour palettes. In the present work a properly combined choice of the confusion points, that are at the basis of these functions, and of the RGB-display primaries gives equiluminant-chromatic palettes, that with the Weberian-lightness define an uniform-colour space suitable for the colour-signal coding. All the transformations are logarithmic. The choice of the RGB primaries can be made with high compatibility with the Ives-Abney-Yule compromise.
Bibliography
[1] C.Oleari, Uniform-Colour Space for Colour Displays, The Journal of Photographic
Science, 43, 24-29 (1995)
[2] C.Oleari: Uniform-Scale-Chromaticity Diagrams by von Kries-Invariant
Logarithmic Transformations for foveal vision, Die Farbe, 43, 17-38 (1997)
[3] C.Oleari, Uniform-scale chromaticity diagrams: opponent-chromatic responses
as logarithms of the cone-activation ratios, Color Research and Application,
in press (1,...,1998)
Restoration of Color Movies by Digital Image Processing
Authors: PD Dr. R. Gschwind, Dr. R. Rosenthaler, Dr. W. Graff,
A. Wittmann Phone: +41-1-6324416
Fax: +41-1-6321210
E-mail:
A system for restoring old, degraded and damaged color motion-picture films using digital image processing is presented. Movie films are based on photographic materials and these are rather unstable compared to other cultural objects. During the next decades thousands of films will deteriorate beyond any hope of restoration. Since photography and motion pictures play an important role in our modern life as document, memory, information carrier, artistic medium etc., photographic material represents an important cultural value. Methods need to be found to reconstruct the colors, and correct other damages like dust and scratches produced by mechanical wear and abrasion.
Restoration of damaged historic motion-picture films by chemical means or by photographic copying is slow and expensive and works only if the degradation has not proceeded too far.
The restoration can be carried out in the digital domain. High resolution digitization and image processing algorithms provide a way to reconstruct accurately and preserve the information in the long term.
Multispectral-Based Color Reproduction Research at the Munsell Color Science Laboratory
Author: Roy S. Berns
Phone: +1-716-475-2230
Fax: +1-716-475-5988
E-mail:
In traditional graphic reproduction, spectral and colorimetric data are used for quality assurance. Characterizing the accuracy between proof and print, determining whether SWOP specifications have been met, and quantifying print variability have all benefited from spectrophotometry and colorimetry. The traditional techniques of image capture, scanning, proofing, and separating do not take advantage of colorimetry and spectrophotometry. Rather, a combination of densitometry and experience is used. Because these systems are closed-loop systems, process measurements (densitometry) and visual adjustments have well served the printing industry producing high-quality graphic reproductions for most applications.
However, for critical color-matching applications such as catalog sales, art-book reproductions, and computer-aided design, the images produced using traditional techniques, although pleasing, are unacceptable with respect to color accuracy. The limitations that lead to these errors have a well-defined theoretical basis and are a result of current hardware and software. The problems are further exacerbated because the field has changed dramatically during the last several years. Open architecture, digital technology, desktop publishing, short-run printing, and a reduction of an experienced workforce further accentuate the limitations of traditional approaches.This has led us to a re-examination of the traditional graphic reproduction paradigm. From this examination, we discover that many of the limitations associated with traditional approaches are due to image capture, the use of SWOP inks, and color separation algorithms based on the concept of photographic masking.
A research and development program has begun that will alleviate the
theoretical limitations associated with traditional techniques. There are
three main phases:
I. Multi-spectral image capture.
II. Spectral-based separation and printing algorithm development.
III. Implementation on press.
IV. Systems integration with data and image archives.
Research has begun on Phases I and II. Noise analyses in multispectral
image acquisition has been completed [1-4]. We have recently installed an
IBM Pro/3000 12-bit colorimetric camera. This camera will be used in conjunction
with either spectroradiometric or low-resolution multispectral digital image
capture to estimate spectral scene information on a pixel basis. The second
phase involves using the spectral data for ink selection followed by multi-ink
printing. Initial research based on linear modeling and canonical corelation
analysis as a tool for ink selection is yielding promising results [5].
Spectral models of halftone printing indicate that analytical models can
be incorporated with the required accuracy [6-8]. This presentation will
describe this new paradigm, summarize recent research results, and consider
implementation opportunities.
[1] P.D. Burns and R.S. Berns, Analysis of multispectral image capture,
Proc. Fourth IS&T/SID Color Imaging Conf., 19-22 (1996).
[2] P.D. Burns and R.S. Berns, Modeling colorimetric error in electronic
image acquisition, Proceedings OSA annual meeting, in press (1997).
[3] P. D. Burns, Image Noise and Quantization in Multispectral Capture of
Color Images, Ph.D. dissertation, Rochester Institute of Technology, 1997.
[4] P.D. Burns and R.S. Berns, Error propagation in color measurement and
imaging, Color Res. Appl., 22, 280-289 (1997).
[5] D.Y. Tzeng, Spectral-based color separation algorithm development for
multiple-ink color reproduction, RIT Center for Imaging Science dissertation
proposal, 1997.
[6] K. Iino and R.S. Berns, Building color management modules using linear
optimization I. desktop color system, J. Imag. Sci. Tech., in press, 1997.
[7] K. Iino and R.S. Berns, Building color management modules using linear
optimization II. prepress system for offset printing, J. Imag. Sci. Tech.,
in press, 1997.
[8] K. Iino and R.S. Berns, A spectral-based model of color printing that
compensates for optical interactions of multiple inks, Proc. International
Association of Colour, Colour97, in press, 1997.
Performance estimation of color printing models
Author: Werner Praefcke
Phone: +49-241-807677
Fax: +49-241-8888196
E-mail:
We evaluate the best achievable performance of three different types of color printers, which are the Neugebauer model, the Yule-Nielsen model and the Kubelka-Munk model. In an extensive optimization we calculate the best primary spectra that lead to a minimal average color error for a set of test spectra under a set of illuminations. The results indicate, that the Yule-Nielsen model, which is distinguished by consideration of the dot gain, yields better results than the plain Neugebauer model. Both are surpassed by the Kubelka-Munk based model.
Introduction: In conventional color printers the color output is controlled by the three printer driving parameters, which are the CMY values defining the density of the three primary colors cyan, magenta and yellow. With these three degrees of freedom it is only possible to match an original color for one specific illumination. Since the reproduced reflectance spectrum needs to comply with the range of the spectral gamut of the printer, it must be expected that it differs from the original spectrum. Making use of the effect of Metamerism it is however possible to achieve an exact match in terms of color values, even though only for one single illumination.
Optimization Approach: Since we can not achieve exact color match for several illuminations we aim for the minimization of the mean color error for a set of illuminants. With this precondition we estimate the best possible performance with respect to different mathematical printer models, which are the Neugebauer model, the Yule-Nielsen model and the Kubelka-Munk model. The basic approach is to optimize the reflectance spectra of the three primaries to minimize the above given mean color error for a set of test spectra.
The procedure consists of two nested optimizations, the outer for the printer primary spectra and the inner for calculating optimal CMY-values, the printer driving parameters. A detailed description of the scheme is found in a recent publication [1].
In extension to the original Neugebauer model we also take the dot gain into account, which expresses itself in the Yule-Nielsen formula. For the Yule-Nielsen factor we choose a value of 2. Furthermore we employ the mathematical model for a dye diffusion thermal transfer (D2T2) printer, which is described in [2]. This continuous tone printing scheme is based on the Kubelka-Munk theory.
We perform the optimization of the contributing primary spectra, which are the reflectance spectra in case of the Neugebauer model and the Yule-Nielsen model, and the absorption spectra for the D2T2 model. The quality measure to be minimized is a composition of the mean squared $\Delta E^*_{94}$ color error, added by a measure of roughness of the spectra. By that means we achieve relatively smooth primary spectra making them more likely to be producible.
Results: Preliminary results are given in Table 1. We compare the mean
and the mean squared color errors of the three models with the errors of
the Neugebauer model using primary reflectances from a Cromalin print proof
process.
Print model |
mean error |
mean squared error |
|---|---|---|
Neugebauer, Cromalin primaries |
1.93 | 7.41 |
Neugebauer, optimized primaries |
0.70 ... 0.73 | 0.78 ... 0.85 |
Yule-Nielsen |
0.55 ... 0.56 | 0.52 ... 0.54 |
Kubelka-Munk |
0.43 ... 0.44 | 0.37 ... 0.39 |
Since successive optimization runs produce slightly different results, we give the range of resulting error values. All three printer models outperform the conventional Neugebauer model based on Cromalin primaries. As an interesting result, the performance with consideration of the dot gain turns out better than with the plain Neugebauer model. This effect corresponds with other observations[3] of enlarged color gamuts when dot gain is taken into account.
Therefore the dot gain effect must be regarded as a benefit to color printing rather than an unwanted imperfection. The Kubelka-Munk based model, which works with a completely different continuous tone color mixing type, leads to color errors of the same magnitude than the halftone models, and even outperforms them both.
Our optimization renders reflectance spectra for printer primaries with the constraints to be restricted to the interval between 0...1, and at the same time to be as smooth as possible. Further work should pay attention to the current technical possibility of producing inks with certain reflectance spectra. These additional constraints could easily be implemented in the optimization process.
References
[1] W. Praefcke. Calculation of best CMY basis functions. In Proc.IS&T/SPIE's
Symposium on Electronic Imaging, volume 3300, accepted for publication,
San Jose, California, USA, January 1998.
[2] R. S. Berns. Spectral modeling of a dye diffusion thermal transfer printer.
J. Electr. Imag., 2(4):359-370, October 1993.
[3] S. Gustavson. The color gamut of halftone reproduction. In Proc.Color
Imaging Conference, pages 80-85, Scottsdale, Arizona, USA, November 1996.Restricted
to the interval between 0...1, and at the
Implementation of a Segmentation Method for Agricultural Fields in Aerial Sequences of Images Based on CSAR Model
Authors: Haijun Chen and Zweitze Houkes
Phone: + 31.53.489 1031
Fax: + 31.53.489 1067
E-mail:
In this paper, we propose the implementation of a segmentation method for agricultural fields observed in aerial sequences of images based on the CSAR (Circular Symmetric Auto-Regressive) Model. One of the main requirements for object recognition and classification is a reliable segmentation. All subsequent interpretation tasks, including object detection, feature extraction, parameter estimation rely heavily on the quality of the segmentation results. The goal of image segmentation is to process the data of an acquired image so as to arrive at a meaningful partitioning of the image plane. The partitioning is meaningful if a correspondence is known to exist between segments on one hand and objects of the scene being imaged on the other hand. Image segmentation methods are based on one or more features. The type of image features of the acquired images can basically be divided into shape, color, texture, etc. We will use texture for the image segmentation. Texture could be defined as a structure composed of a large number of more or less ordered similar elements, called texels, of patterns without one of these drawing special attention. A texel contains several pixels, whose placement could be periodic, quasi-periodic or random. Texture provides essential structural information of regions in an image and is therefore a valuable feature for segmentation. The task of texture segmentation is to segment the images into a number of regions such that each region differs from neighboring regions in the textural properties. Texture based segmentation methods can roughly be categorized as feature based and model based. In feature based methods, some characteristics of textures are chosen. Regions in which these characteristics are relatively constant are sought. Alternatively, one can look for the boundaries between these regions. Model based methods hypothesize the underlying process for textures and segment using certain parameters of these processes. The two major characteristics of a texture are its coarseness and directionality, which are used in our model. Since model parameters are used as texture features, model based methods could be considered as a subclass of feature-based methods. In this paper, the Circular Symmetric Auto-Regressive model is used to describe the texels for the segmentation of images. It is based on the well-known fact that each pixel in an image is linearly dependent on its neighbors and the property of a texel is assumed to have piece-wise performance. The image sequences are acquired by a video camera from an aeroplane, which moves linearly over the scene. The objects in the scenes, being considered in this paper, are agricultural fields. The image segmentation of the agricultural fields to be distinguished is carried by the Circular Symmetric Auto-Regressive model. Based on reliable segmentation of image sequences, the recognition and the classification for surface texture of vegetation from aerial sequences of images are realized.
Biography: Haijun CHEN received the B.Sc., M.Sc. and Ph.D. degrees in Electrical Engineering, from Harbin Institute of Technology (HIT), Harbin, China, in 1986, 1989 and 1995, respectively. During his Ph.D. candidate period, he has been Eindhoven University of Technology, Dept. E. E. and Vrije Universiteit Brussels, Dept. ELEC, as a Visiting Scholar. He is currently a post-doctoral at Laboratory for Measurement and Instrumentation, Department of Control Systems and Computer Engineering (BSC), University of Twente, The Netherlands. His current research interests are the areas of image processing, system identification, model based measurement, digital signal processing. Zweitze HOUKES was born in Utrecht, The Netherlands, in 1939. He received his M.E. in electrical engineering from Delft University of Technology, The Netherlands, in 1963. In 1968 he joined the Electrical Engineering Dept. of the University of Twente, where he was engaged in research on Control Engineering. Since 1984 he is an associate professor of Computer Vision. He is member of the International Association for Pattern Recognition, the Dutch Association for Pattern Recognition and Image Processing and the Dutch Foundation for Measurement and Control.
OPTIMAL CHANGES IN THE PREPRESS PROCESS A model for evaluation of state and direction of productivity and quality improvements.
Authors: Mats Lindgren and Henrik Kihlberg
Phone: +46-8-453 57 25
Fax: +46-8-453 57 57
E-mail:
The demands changes with the customers marketplace which makes it crucial for the GASP (Graphic Arts Service Providers) of today and those of tomorrow to be able to change their services with it. The production tools are getting more standardized and similar for all of them. Intelligent tools are developed in a rapidly growing speed which results in possibilities to automate the processes. The factors of success for the actors of today and tomorrow are the ability to change and understand the customers market. The market demands shorter delivery times and lower costs. The total number of printed editions are decreasing with a growing numbers of pages and images. The customers requires higher quality with the ability to control and predict the end result.
Case studies, interviews and workshops have been carried out at commercial printing companies, prepress houses, image bureaus, advertising agencies and digital photographers in Sweden. A major part of the research have been focusing on the digital image process at nine GASP which all have a prepress workflow. The analysis has resulted in a thorough knowledge of the production workflow and parameters to measure the productivity and the quality. A model for evaluation of changes is presented, with measurable values for productivity and quality. The model can be used to map and compare the states the GASP are in and or be used to evaluate if the changes are optimal.
Key words: Model, Productivity, Quality, Knowledge, Communication, Throughput, Workflow, Process, Activities, Customer, Image and Pages
Testing of Colour Management Systems
Author: Walter F. Steiger
Phone: +41-71-2747-666
Fax: +41-71-2747-663
E-mail:
CMS-Background (seperate) A cms study, which will be finished by the end of this year, compares reproductions made with ICC-profiles from four different color management softwares. Materials and Methods:The CMS-reproductions were compared to a "standard" reproduction, 84 of the patches in the ISO 12641 input target (former IT.8.7x) were measured and compared to target values of the standard reproduction. Color differences were calculated in terms of delta E values and the reproductions were evaluated in terms of lightness, hue and chroma rendering.
The reproductions were made with ICC-profiles created for various input (scanner, digital camera) and output devices (color laser printer, dye-sublimation printer, offset-proof).
Results and discussion: Within December 97
Key Words: Color Rendering, CMS Workflow, ICC Limitations, Gamut Mapping
DNC applied to colour object recognition with tolerance to slight shape and colour variations.
Authors: Elisabet Pérez, Mar¡a Sagrario Mill n
and Katarzyna Chalasinska-Macukow
Phone: 34-3-739 83 39
Fax: 34-3-739 83 01
E-mail:
Dual Nonlinear Correlation (DNC) has been recently introduced as a general operation in optical pattern recognition involving well-known linear and nonlinear filtering methods (classical matched filter-CMF, phase-only filter-POF, inverse filter-IF, pure phase correlation-PPC, etc.). Not only does the DNC include these previous filtering proposals, but also allows to apply new non-symmetric operators to both the analysed scene channel and to the reference target channel.
The implementation of the DNC is carried out in a sole filterless optoelectronic processor based on a two-step joint transform correlator assisted by computer. Simply by changing the values of two parameters it is possible to introduce a large variety of nonlinearities, including the previous standard filtering methods as well as intermediate cases. Experimental conditions such as dynamic range of the CCD camera, quantization of the acquired image and grey level quantization of images displayed in the spatial light modulator placed in the optoelectronic processor may considerably alter the predicted theoretical results of a given situation of target recognition.
In this work we present an application of the DNC to colour pattern recognition taking into account experimental constraints. In colour pattern recognition, correlation techniques that apply standard filtering methods frequently lead to systems with a discrimination capability too rigid for a number of situations. A relaxation of the threshold applied to the final correlation intensity distribution may solve a requirement of tolerance but it also entails the risk of increasing false alarms that originally were avoided. In our paper, we aim to recognise a colour object even though it has slight variations in colour and shape. For this case, we consider the recognition of colour letters (belonging for instance to a sign, advertisement, logotype, etc.) with certain tolerance to shape and colour variations. We consider irrelevant small changes in letters shape as well as colour variations within a given gamut of hue.
When using conventional filtering processes, the recognition result is limited by the method performance. On the contrary, by using a proper DNC, which we control through two parameters, we have achieved a colour pattern recognition system with the required tolerance in its discrimination capability.
Key words: Nonlinear processing, colour pattern recognition, joint transform correlator.
Biography: Elisabet Pérez received her BS degree in Physics from the Autonomous University of Barcelona in 1993 and MS degree in Applied Optics from the Polytechnic University of Catalunya (UPC) in 1995. She is currently finishing her PhD in the field of colour pattern recognition in the UPC, where she is an assistant professor. Her main research interests are optical colour pattern recognition, liquid crystal devices and object classification.
Color Appearance : Effects of Texture and Relief
Authors: Philippe LONGERE and Alain TREMEAU.
E-mail:
Introduction: The color appearance of a surface depends on various parameters which include global color, local texture and relief of the surface that is observed. As we wish to describe a surface as it appears to human observer, the camera needs to be calibrated in order to fit, as good as possible, the properties of the CIE64 observer [1].
Calibrating the camera can be considered as finding the best mapping between the RGB values of the acquired material on the XYZ space. It is commonly assumed that this mapping is linear and depends on the chromaticity coordinates of color primaries and the white point.
Methods: Using the McBeth Color Checker as the first step in the acquisition program, we can calculate the transfer matrix to XYZ space that matches the real CIE values of the color patches. This is a two step process that includes luminance balancing and space transfer matrix numerical computing. Once this matrix is obtained, we estimate the average difference between real and acquired color patches to qualify the accuracy of our computation [2].
Using the approximate spectral response of the camera, we can use the transfer matrix to compute the "theoretical matching curves" of the camera, the way the camera behaves as a human observer. We can now use the acquired images to characterize the appearance of the surface. Using the camera as a photometer, we can have a global information about the color of the material, and then try to find out information about the spectral properties of the material. This can be accurately done with material such as Munsell chips [3], or the McBeth checker, as it has been proved that their reflectance depends only on a small number of parameters. Dealing with other materials, this can be a tough problem because reflectance is closely linked with the physical properties of the surface.
But a photometer can only give an integrated value of the color of a surface, and in order to give an accurate characterization, we introduced a description of the texture of the material. Texture can be qualified with a lot of statistical parameters, we chose to use a few parameters calculated with the color co-occurrence matrix [4] . This makes it possible to modulate the color information by a description of the texture that composes this color, such as inertia, or entropy [5].
When trying to qualify a surface, one could try to describe its roughness, or smoothness. This appears on image under the varying of orientation of the lighting, as a rising light shows the roughness. Relief can be calculated as "the apparition of texture with a rising light", though we take into account the variation of the above texture in order to quantify the relief of the material under study.
Results: This study offers to describe the color appearance of a surface as a combination of color measurement, texture description and relief. Colorimetric values come with texture description and relief estimation that give further description of the appearance of the material.
References
[1] Strachan N.J.C., Nesvadba P. and Allen A.R.(1990), Calibration of a
video camera digitising system in the CIE Luv colour space, Pattern recognition
Letters 11(11) pp771-777.
[2] Schettini R., Barolo B.and Boldrin E.(1995), Colorimetric calibration
of color scanners by back-propagation, Pattern recognition Letters 16 pp1051-1056.
[3] Cohen J.(1964), Dependency of the spectral curves of the Munsell color
chips, Psychonomic Science 1: 369-370.
[4] Trémeau A., Colantoni P. and Laget B. (1996), On color segmentation
guided by the cooccurrence matrix, OSA Annual conference on Optics & Imaging
in the Information Age, Rochester pp30-38.
[5] Trémeau A., Bousigue J. and Laget B. (1996), Cooccurrence shape
descritors applied to texture classification and segmentation, SPIE Proceedings,
San Jose, Vol 2665 pp135-147.
Biography : LONGERE Philippe is a PhD student at the Institut de l'ingénierie de la Vision at the Université Jean Monnet at Saint-Etienne, France. He's currently in his second year of doctoral study on image processing. His main interests concern color appearance and recovery of spectral properties. TREMEAU Alain received a PhD in Image Analysis in 1993 from the University of Saint-Etienne. In 1994 he joined the University as a lecturer and as a researcher. His research interests are in the area of mathematical imaging and color vision with reference to human vision and perception. He is also very interested in color metric analysis with regard to color appearance and rendering measurement.
A tutorial on colorimetry, based on hypertext and multimedia
Authors: Cecilia Sik-L nyi and J nos Schanda,
Phone: +36 1 176 5394
E-mail:
Colour is an essential part of information handling. Computer science (informatics) students usually do not get a formal education in colour science, but show interest in self study of the subject. Hypertext and multimedia presentations can be used with advantage in such situations. A CD-ROM based self-study manual has been prepared for informatics students to teach them colour science, where the tutorial is built up by starting with some general errors, recommendation for good practice and links guide the student to more profound understanding of the underlying colorimetric knowledge and functioning of display hardware systems.
Display white point characterisation
Authors: Cecilia Sik-L nyi, Zolt n Györffy, Peter Bodrogi,
Janos Schanda
Phone: +36 1 176 5394
E-mail:
The white point characterisation of display images reaches new interest due to the widespread application of computer image processing and electronic printing. Some years ago it was the general consensus that the maximum composite signal of the red, green and blue channels should be used as the white point of the display. In characterising this white point the concept of correlated colour temperature is used, based on the outdated (u,v)-diagram.
An experiment was conducted to investigate the appropriateness of the concept of the correlated colour temperature. Experiments showed that a colour difference in CIELAB colour space is a better descriptor of correlated colour temperature then the traditional method. It is recommended that for evaluating the correlated colour temperature of computer screens and for comparing the white point of the screen with that of input pictures and printed outputs the new concept of minimum colour difference in CIELAB colour space be used.
References:
[1] Carter R C, Colour Forum: CIE L*u*v* color-difference equations for
self-luminous displays, COLOR 1993, Vol. 8, No. 4, 252-253
[2] Kelly K L; Lines of constant correlated color temperature based on MacAdam's
(u,v) uniform chromaticity transformation of the CIE diagram,JOSA 1963,
Vol. 53, No. 8, 999-1002.
[3] Schanda, J., Correlated colour temperature and the colour-difference
formula, Proc. AIC COLOR'77, Troy, N.Y., 292-294, 1977.
[4] Bodrogi, P., Schanda, J., Testing the calibration model of colour CRT
monitors. A method to characterise the extent of spatial and channel interdependence,Displays
16/3 123 - 133 1995.
Comparison of color median filters
Authors: Philippe COLANTONI, Alain TREMEAU, Bernard LAGET
Phone: +33 477923030
Fax: 33 477923039
E-mail:
1. Introduction: The nonlinear image processing techniques have known a increasing development in last years. Noise filtering and enhancement of vectorial images and of multispectral remote-sensing data are examples of applications. In our case we work with 3 dimensional signal : the color image. We propose in this work to study two different methods in order to realize a median filter of a color image.
2 Methods: The use of the median operator in image processing was introduced by Tukey [1]. The median filter performs a nonlinear filtering operation where a window moves over a signal, and, at each point, the median value of the data within the window is taken as the output. We propose here two methods to realize this operation. These methods are based on the establisment of an order within a 3 dimentional data set. This data set localy defined at pixel (x,y) , according to a pxp neighborhood mask. We have seen previously that we need to define an order relation between all the points of this data set to realize the median filter. We propose here two methods to establish this order relation.
2.1 Using the local principal component. The transform proposed here is based on the projection of the colors on the first principal axe of the data set. According to this transformation we have for each neighborhood mask one dimensional signal.
2.2 Using bit mixing The transform proposed here is based on the representation of each color component in the binary mode[2]. In other words, this mapping that "projects" the color pixels from their three dimensional base RGB onto R encodes each pixel by mixing up the 3*8 available bits together, taking alternatively one bit from the red, the green and the blue component to build a 24 bits long integer [3]. Now it's easy to sort the 1 dimensional data and to find the median of the data set. We have compared these two methods with an other based on color euclidian distance [4].
3. Results and discussion: The results are very similar in the two cases but the computation time is different. The sort process is the same for each method, the diffence is in the transformation process. In fact the principal compoment transform is time consuming because it implies a lot of computation for each pixel, the bit mixing is then instantaneous. The bit mixing works fine because it is an approximation of the color projection on the luminance axe. But the principal compoment works for all color distribution in a color image.
References
[1] J.M. Tukey, ® Nonlinear (nonsuperposable) methods for smoothing
data ¯, in Congr. EASCON, p.673, 1974.
[2] C. Gu, ®Multivalued Morphology and Segmentation based Coding ¯,
Ph. D. Thesis of the Ecole Polytechnique Federale de Lausanne (EPFL), 1995
[3] J. Chanussot and P. Lambert, ®Bit Mixing Paradigm for Multivalued
Morphological Filters ¯, in Congr. IPA 97, Volume 2, p 804-808
[4] A. J. Bardos and S.J. Sangwine, ®Selective Vector Median Filtering
of Colour Images ¯, in Congr. IPA 97, Volume 2, p 708-711
Key words : median filter , color, bit mixing , principal compoment.
Biography: Philippe Colantoni is a PhD student in the Vision and Image Analysis group managed by Professor Bernard Laget (University Jean Monnet of Saint-Etienne, France). His research interests are in field of color image segmentation and human vision.
One-dimensional Dithering
Author: Bob Ulichney
Phone: +1-617-692-7633
Fax: +1-617-692-7640
E-mail:
A real-time inverse dithering system is proposed for video display that operates on only the current scan line. To optimize overall display quality, a corresponding one-dimensional ordered dither array is sought. This paper describes a one-dimensional recursive tessellation algorithm. A serendipitous implementation involves a simple bit-reversal of the horizontal pixel address. To optimize two-dimensional homogeneity, the 1-D array is phase adjusted in the vertical direction. A scheme for selecting candidate phase vectors is also presented. The recursive tessellation algorithm is generalized to identify equivalence classes arrays that share the same homogeneity property but have different ordering.
Spectral characterisation of electronic cameras
Authors: Jon Yngve Hardeberg, Hans Brettel, and Francis
Schmitt Phone: +33 1 45 81 8145
Fax: +33 1 45 81 37 94
E-mail:
It is current practice to use standard colour targets such as the IT8.7/2 and to apply analytical models for the mapping of the input device data into a standardised device independent colour space. The mapping function is typically obtained by minimising the mean square error of a set of measurement points by polynomial regression, see e.g. [1-4]. An important limitation of such methods is that individual characterisation data have to be obtained for each type of input media, the failure to do this resulting in considerable errors due to metamerism. Furthermore, these methods are purely mathematic, they do not at all take into account the physical characteristics of the colour image input device.
For a more complete characterisation, the knowledge of the physical characteristics of the different optical and electrooptical components which are involved in the image conversion process would be desirable.
Firstly, the eventual non-linearities of the electrooptical components have to be characterised and corrected for. Then, the camera response can be expressed by a linear model as the wavelength integral of the product of the spectral sensitivity of the sensor, the spectral transmittance of the optical path, the spectral reflectance of the object, and the spectral radiance of the illumination, plus a noise factor, see e.g. [4-6]. By sampling uniformly the spectra, we may formulate the camera model in terms of algebraic matrix operations.
The spectral characterisation consists in estimating the different spectral characteristics of the sensor, the optics and the illumination, or eventually, the joint characteristics of these elements.Theoretically, in the absence of noise, this characterisation may beobtained to the desired spectral accuracy by measuring a set of P samples of known linearly independent spectral reflectances and by inverting the resulting system of linear equations, for example by using the Moore-Penrose pseudoinverse. Thus the spectral characteristics can be determined for a sampling rate of N wavelengths, N <= P [7]. However, in the presence of noise, and due to the fact that typically only a small number of sample spectra are independent, this system inversion becomes hazardous.
A common solution for this kind of inverse problem is to use a method based on a singular value decomposition (SVD) and to use only those components whose singular values are greater than a threshold value. This solution is often referred to as the principal eigenvector (PE) solution [5,6]. Unfortunately, by using such a method, the resulting principal vectors are linear combinations of all the sample spectra, i.e. it is not obvious which of the sample spectra are really relevant to the characterisation.
Here we present an alternative method in which we start from a small number of sample spectra which are chosen according to their spectral variance. Additional sample spectra are added in order to maximise the volume defined by the vector space spanned by these spectra, calculated as the determinant of all the chosen sample spectra. This method yields results which allow a higher spectral resolution of the reconstructed spectral sensitivity functions compared to the PE method.
We have investigated the methods using different sets of reflectance
samples, such as Munsell colours and oil painting pigments, both by simulations
and by applications onto real colour input devices. The proposed method
is applicable not only to electronic cameras but also to other electronic
image input devices such as desktop scanners. The method may also be applied
to the task of designing optimized colour targets.
[1] Roy S. Berns, (1993). Colorimetric characterization of Sharp JX610 desktop
scanner. Technical report, Munsell Color Science Laboratory.
[2] Tony Johnson, (1996). Methods for characterizing colour scanners and
digital cameras, Displays, 14(4), 183-191.
[3] Jon Yngve Hardeberg, Francis Schmitt, Ingeborg Tastl, Hans Brettel and
Jean-Pierre Crettez, (1996). Color Management for Color Facsimile. Proceedings
of IS&T and SID's 4th Color Imaging Conference: Color Science, Systems and
Applications, pages 108-113, Scottsdale, Arizona.
[4] Henry R. Kang, (1997). Color Technology for Electronic Imaging Devices,
SPIE Optical Engineering Press
[5] Joyce E. Farrell and Brian A. Wandell, (1993). Scanner linearity. Journal
of Electronic Imaging, 2(3),225-230.
[6] Gaurav Sharma and H. Joel Trussell, (1996). Set theoretic estimation
in color scanner characterization, Journal of Electronic Imaging, 5, 479-489.
[7] Henri Ma?tre, Francis Schmitt, Jean-Pierre Crettez, Yifeng Wu and Jon
Yngve Hardeberg, (1996). Spectrophotometric Image Analysis of Fine Art Paintings,
Proceedings of IS&T and SID's 4th Color Imaging Conference: Color Science,
Systems and Applications, pages 50-53, Scottsdale, Arizona.
Biography: Jon Yngve Hardeberg received his sivilingenior (M.Sc) degree in 1995 from the Norwegian Institute of Technology, Trondheim, Norway. He is actually pursuing his Ph.D studies at the Ecole Nationale Supérieure des Télécommunications, Paris, France. His research concerns color acquisition and reproduction, with applications in facsimile, fine-art paintings, and multispectral imaging.
Adaptive compression of still images: automating the choice of
algorithm and parameters Authors: F. Gilbert, F. Moutard
E-Mail:
This paper presents a new adaptive compression scheme for continuous-tone still images with two main benefits: 1ø/ it allows to take advantage of a set of algorithms (such as JPEG, wavelets, fractals, ...) by switching between them according to the image content and target compression ratio; 2ø/ it tunes automatically the algorithm-specific parameters (e.g. JPEG quantization factor) in order to meet the user compression objective (e.g. a target compression ratio). A performance evaluation test indicate the following: 1ø/ our adaptive coder can succeed with probability ( 72% in selecting automatically the best technique (among JPEG, a wavelet algorithm, and a fractal one); 2ø/ the difference between the obtained compression ratio and the target is lower than 10%; 3ø/ the computing overhead due to algorithm selection and parameter choice is negligible. This adaptive compression technique has been proposed to the ISO standardisation committee currently working on the definition of the future "JPEG-2000" standard.
Adaptive Color Correction based on Objects Color Classification
Authors: Hiroaki Kotera, Teturo Morimoto, Nobuyuki Yasue, and
Ryoichi
Saito Phone: +81-43-290-3486
Fax: +81-43-290-3490
E-mail:
A natural color image is composed of different colored objects distributed in space. Each object has its own surface colors distinguished from others which are often aggregated in a specified area of three dimensional color space.
This paper reports an adaptive color correction strategy to make match the reproduced colors to the originals depending on the objects' color distributions in color space. Here both an original image and the once printed image on a hardcopy are captured through digital camera or scanner, then the color differences between two images are analyzed and corrected.
First, a color classification algorithm is discussed based on statistical color distance index. RGB color components are transformed to CIELAB space and the clustered LAB values are analyzed and classified into several key colors dominant in its image. Here color distance indices such as Euclid, Mahalanobis, and Chernoff distances are used to classify the objects colors in three dimensional color space. The LAB components in each color cluster are approximately distributed in an ellipsoid.
Secondly, the color distributions in each classified cluster corresponding to the original and the printed image are projected onto principal components space by Hotteling transform from LAB color vector x to y as
color vector, m=mean vector Here, a new vector y is obtained in terms of matrix A whose rows are formed from the eigen vectors of covariance matrix Cx given by Cx=E{(x-m)(x-m)T} ; T denotes transpose. The covariance matrix Cy of color vector y is given by A and Cx as follows.
There, Cy is a diagonal matrix whose diagonal elements are the eigen vectors of Cx. As a result, the transformed color vectors in original image and printed image are mapped to the same principal components space and compared each other.
Finally, the color correction matrices are determined to make match the ellipsoids formed by principal components in printed image to those of original image in individual color cluster and applied to correct the pixel data in the original.
In the coming paper, the experimental data and estimations will be introduced.
Development and Standardization of a Spectral Characteristics DataBase (SCDB) for Evaluating Color Reproduction in Image Input Devices
Author: Dr. Johji Tajima
Phone: +81-44-856-2145
Fax: +81-44-856-2236
E-mail:
[Background] It is generally agreed that spectral sensitivities in the color sensors employed by image input devices need to satisfy the Luther condition; i.e. these sensitivities should be expressed by linear combinations of color matching functions or human cone sensitivities. For evaluating the degree to which the Luther condition is satisfied, Neugebauer's quality factor[1] or the measure of goodness by Vora and Trussell[2] has been proposed. However, they assume that object spectral characteristics (transmittances and reflectances) are uniformly distributed on entire multi-dimensional spectral vector space. Because this assumption does not hold in practice, these evaluation methods cannot be used effectively. The Luther condition is only a sufficient, not a necessary, condition for good color reproduction[3]. In practical sensor sensitivity design for image input devices, real color samples (e.g. Macbeth color charts) are used to evaluate color reproduction quality, but no standardized set exists for this purpose, and evaluation results simply depend on which color samples have been used.
[Development] It was in response to this situation that the development of the SCDB was first proposed. The database is to be designed with the following two features:
The separate categories for the SCDB are:
Category a is for application to film scanners; categories a, b and c are for color scanners; categories d f and g are especially important for digital and video cameras, though these devices can input characteristics from any other category as well; category e is to be an important element in digital archives. Having collected spectral characteristics according to the categorization, we are now conducting a statistical analysis, and its results will be presented at the symposium. For color reproduction evaluation purposes, it is important that:
[Future plan] The database is to be published as a technical report by the JSA and to be used as a standard database in Japan. It is also hoped that it might become the international standard for color reproduction evaluation in image input devices. For the purpose, international collaboration on collection and evaluation is inevitable, since sample color distribution in characteristic categories b through e depends largely on geographical location.
[References]
[1] Neugebauer, "Quality factor for filters whose spectral transmittances
are different from color mixture curves, and its application to color photography",
J. Opt. Soc. Am. Vol. 46, No.l0, pp.821-824 (1956)
[2] Vora and Trussell, "Measures of goodness of a set of color-scanning
filters", J. Opt. Soc. Am. A, Vol.10, No.7, pp.1499-1503 (1993)
[3] Tajima, "New Quality Measures for a Set of Color Sensors-Weighted
Quality Factor, Spectral Characteristic Restorability Index and Color Reproducibility
Index", Fourth Color Imaging Conference, pp.25-28 (1996)
Analysis of the compensating mechanism of a colour CCD camera for changes of illuminant light
Authors: Montse Corbal n, Mar¡a S. Mill n, Mar¡a
J. Yzuel*.
Fax: 34-3-739 83 01
E-mail:
Key words: Colour CCD camera, RGB video acquisition, white balance, CIELAB space
When the illuminant source changes during the image acquisition the colour content of a digital colour image changes as well. Colour cameras usually contain a device called white balance that allows to capture a white reference without predominant colours. This adjustment is performed when the light source changes in order to compensate for possible variations in its spectral emittance. If the compensation is not complete, then variations in the RGB values of a colour image are usually registered. We have analysed the compensating mechanism of white balance in a colour CCD camera. For this study we have tested a Sony DXC-930P camera which may represent the general performance of a colour camera with three CCD sensors. We have captured a hundred colour sheets which cover the visible spectrum under the separate illumination of three common lights (fluorescent, incandescent and daylight).
From the analysis of the variations in the R, G, B values obtained for each pair of illuminants, we conclude that, in this case, a change of illuminant from daylight to incandescent (or vice versa) is better compensated than a change of illuminant in which fluorescent light is involved (i.e. from daylight or incandescent to fluorescent, or vice versa). This effect is more intense in the red channel than in the green and blue ones.
Taking into account the great importance of CIELAB coordinates in industrial applications and the increasing use of colour CCD cameras in tasks of colourimetric inspection, we have developed a method to obtain the coordinates L*, C*, h*, (Luminance, Chroma and Hue) in the CIELAB space from the RGB components captured by camera. The method is based on a linear transformation. The matrix of the transformation is derived from a least square minimum fitting. Colourimetic distances (chromaticity differences) are calculated between the (C*, h*) points corresponding to each colour sample illuminated separately by the three sources of light (with white balance). The compensating mechanism of white balance is again analysed in terms of chromaticity differences expressed in the CIELAB space.
Biography: Montse Corbal n received her BS and MS degrees in Physics from the Autonomous University of Barcelona in 1990 and 1993, respectively, and her PhD degree in Physics from the Polytechnical University of Catalonia (UPC) in 1997. Since 1991 she has been an assistant professor in the Dep. of Physics and Nuclear Engineering of the UPC. Her fields of research involve pattern recognition, colour and optical image processing. She is a member of the Spanish Society of Optics and the European Optical Society.
"Automatic Out of Range Colour Picture Recognition in Terms of Optical Density using Digital Techniques"
Author: Genevieve Dardier
Phone: +33 1 45 81 70 89
Fax: +33 1 45 81 37 94
E-mail:
This paper, "Automatic Out of Range Colour Photo Recognition in Terms of Optical Density using Digital Techniques", addresses digital techniques used to identify automatically those colour photographs whose optical densities fall outside acceptable norms for our visual system.
Direct applications include automated high speed colour pictures printing devices (such as mini-laboratories) where appropriated algorithms associated with colour image restauration bring increase in the number of better quality pictures. Secondly, there are applications in the area of the "image rich users interface" where new tools for colour pictures ehancement and restoration are available to the interface users. Colour prints falling outside the acceptable norms of optical densities for use in the automated high speed mini-laboratories are caused by badely taken pictures, but not recognized such as during the printing process; regarding the case of the rich image users interface using badely taken photographs, new software is needed to isolate cases of pictures before proceeding successfully to colour image restoration.
Psychology and language science have undoubtly helped computers understand and interpollate texts written in plain language; in this case, a photographic approach combined with various solid fast algorithms for low and more detailled vision induces specific solutions to the delicate problem of machine recognition.
This paper proposes solutions to the recognition problem using methods that single out out of range colour pictures in terms of optical densities. Targets are of two different kinds and indicate the need for different recognition algorithms and latter different restoration methods (not mentioned in this paper). First, the investigation focusses on colour pictures showing high contrast in optical densities such as pictures taken with flash and pictures taken against the light (indoor and outdoor); it is proposed to identify those pictures through optical density histogramm recognition, in particular estimating histogramm parameters which determine mixture densities.
Requirements regarding colour matching between used scanning devices and the visualisation environement are taken into account before proceeding to any algorithm computation.
Results of Maximum Likelihood, E-M Algorithm [1], Moment estimation using Prony's Method [2], Multiscale Analysis [3] and Marquardt fast algorithm [4] applied to the histogramm recognition problem have been compared in terms of efficiency and speed for both low and more detailled vision. Mixture densities parameter estimation applied to the problem of high contrast picture recognition shows good results; in the perticular case of low vision, Marquart fast algorithm combines efficiency and speed.
Secondly, attention is focussed on colour pictures showing an optical
density gradient in a given direction. Straitforward solutions such as plan
fitting are compared and analysed with a new " Three steps Median Based
Estimator" method [5] applied to the two dimensional case. Investigations
have brought interesting results when comparing low and higher vision in
terms of depth recognition on pictures.
[1] "Mixture densities, maximum likelihood and the EM algorithm",
1984, SIAM Review, p 195-239, R. A. REDNER and H. F. WALKER.
[2] "Estimating Components of Univariate Gaussian Mixtures Using Prony's
Method", January 1987, IEEE Transactions of pattern analysis and machine
intelligence, vol.1, H. DERIN.
[3] " Analysis of Multi-Image Magnetic Resonance Acquisitions for Segmentation
and Quantification of cerebral Pathologies", PhD-Thesis, ENST, March
1997, L. AURDAL.
[4] "An Algorithm for Least-Squares Estimation of Nonlinear Parameters",
J. Soc. Ind. Appl. Math., Vol 11, 1963, D. W. MARQUARDT.
[5] "Design of dense disparity map for building recognition in aerial
images", submitted to RFIA AFCET ,Clermont-Ferrand, 1998, S. GIRARD,
P. GUERIN, H. MAITRE, M. ROUX.
Key words: Colour image analysis, Image quality, Mixture density parameter estimation, Median Based Estimator (MBE).
Biography: Genevieve Dardier, Dipl-engineer (M.Sc) in physics from the Federal Polytechnical school of Lausanne (ETH-Lausanne, Switzerland) is a PhD student at the Ecole Nationale Supérieure des Télécommunications, Paris, France, in the vision laboratory. Her Advisor is Professor H. Maitre and she works actually on the recognition et restoration of badly taken colour photographs.
Colour Based Image Retrieval considering Spatial Information
Authors: Ingeborg Tastl, Christian Wolf
E-mail:
The amount of images, which are produced every day has increased dramatically within the last decade. Visual information comes from satellites, medical systems, scientific experiments, from visualization software and all kinds of image capturing systems. In order to enable an efficient usage of all this information image/video databases are indispensable and systems using content-based retrieval techniques are urgently needed. Instead of using only keywords to describe the visual properties of an image, various features can be extracted form images via image-analysis techniques, and used for comparison purposes during the retrieval process. Features are defined as computable properties of an image (or part of an image). They can also be aggregated to domain specific, higher level, more abstract image descriptions. One of the low-level features is colour.
The classical approach of using colour features only, proved to be insufficient for image retrieval purposes. Thus, combinations with other image properties have to be considered. In this paper an approach using colour information in conjunction with information about spatial arrangements of colours within an image is described. The approach consists of three different steps:
The first part is realized by creating a 3D histogram in the CIELUV colour space from the original image and performing a graph based clustering. Each cluster in the feature space is represented by one colour tripel, which is backprojected into the image. Now the original clustering can be refined with the help of spatial information: Pixels with colour values belonging to a certain cluster may be spread over the whole image. With the help of a connected component analysis, all those pixels belonging to spatial connected regions can be detected. Calculating new 3D histograms of the various regions of a cluster and critically evaluating them can result in the creation of new clusters. Thus the original cluster may be subdivided into two or more new clusters because of the spatial information. Finally representatives for each cluster are calculated and stored in nodes of a graph.
In the second part the final clusters are backprojected into the original image, the adjacencies between the clusters are calculated and represented through edges in the above mentioned graph. Details, including images which visualize the different steps of the approach will be given in the final paper.
The specification of a measure, which describes the amount of similarity between two graphs will be described in the third part. Color distances between corresponding nodes of two graphs, differences in the existence of edges, and differences in the relative amount of pixels being represented by corresponding nodes, are considered in the calculation of the similarity of two graphs representing two images. The question whether one image is just a part of another image can also be answered with the similarity measure.
A number of concrete examples will demonstrate the advantages of the new approach compared to histogram based colour retrieval methods.
The diamond ink jet
Authors: E.P. Hofer, S. Aus der Wiesche, C. Rembe, J. Patzer
Phone: +49-7315026300
Fax: +49-7315026301
E-mail:
A novel thermal ink jet heater based on diamond will be presented. In contrast to the conventional actuator systems passivation layers are not necessary which reduces the manufacturing cost and increases the thermal efficiency of the micro heater. In addition, the excellent mechanical properties of diamond guarantee protection against cavitation damages. In the following the theoretical and experimental development process from first ideas up to the final prototype of the diamond ink jet is described. The new thermal ink jet heater has been tested successfully and its advantages are demonstrated by experiments using high speed cinematography.
Investigations on nonreproducible Phenomena in thermal Ink Jet with real high speed cine photomicrography
Authors: C. Rembe, S. Aus der Wiesche, M. Beuten, E.P. Hofer
Phone: +49-7315026329
Fax: +49-7315026301
E-mail:
In the ink jet printer industry the stroboscopic visualisation method is a standard tool for the characterisation of printheads. However, this method fails for thermal ink jets in case of the existence of satellite droplets which are very critical considering print quality. Detailed studies have shown that the phenomenon of satellite droplets is a nonreproducible dynamic process.
Real high speed cine photomicrography forms the basis of a new test setup which allows the visualisation of such highly dynamic nonreproducible phenomena. This new setup has been used to study the ejection, the free flight, and the impact of droplets of an ink jet on the print media under real printing conditions.
Binary Tree Context Modeling of Halftone Images using a Fast Adaptive Template Selection Scheme
Authors: K. Denecker, P. De Neve and I. Lemahieu
Phone: +32-9-264.89.08
Fax: ++32-9-264.35.94
E-mail:
Reproduction of a continuous-tone image on a binary output medium like a printer, requires halftoning. Classical halftoning, which is by far the most popular type of halftoning, replaces the contone pixels by dots of varying sizes on a rotated square grid; each of these dots is composed of much smaller laser spots. Stochastic halftoning on the other hand, replaces the pixels by fixed-sized dots on varying locations in a semi-random way.
Recently, new applications have arised where lossless halftone image compression can be useful for increasing transmission speed and lowering storage costs. In personalized printing, many copies are produced which differ only in a small region; rehalftoning the entire page would be a waste of computer power if the halftones could be stored efficiently. In printing on demand, paper copies are only produced the moment they are actually needed; otherwise, all the halftone data is stored digitally on a server which could be connected to the Internet.
The current lossless bilevel image compression standard JBIG [1] combines finite-context modeling and resolution reduction and uses the QM-coder [2] as an entropy coder. In the highest resolution level, it uses the 10 closest already coded pixels as a context template. In [3], we have shown that this is by far not optimal for classical halftone images. By constructing an image-adaptive context template based on the auto-correlation function and by using a higher-order finite context model, we have obtained compression ratios which are about 30% to 65% higher than those achieved by JBIG.
However, using a predefined context template like JBIG does, binary tree context modeling achieves compression performance than finite-context modeling [4,5]. Whereas the statistical information of the image can be viewed as an array indexed by all possible contexts of a given length in the case of finite-context modeling, binary tree context modeling uses better-chosen contexts of variable lengths. Every node of the tree contains the statistical information corresponding with its context; the depth of a node determines the length of the context. During the tree growing stage, new nodes are added to the contexts which arise most frequently: this assures that the contexts are chosen well.
In this paper, we extend the use of the proposed adaptive context template selection scheme to binary tree context modeling. We use an existing tree modeling scheme as well as some new variants which differ by their node creation (either one or two child nodes are created at a time), by their inheritance mechanism (a new node's statistical information either stems from its father or is initialized to zero) and by their statistical significance criterium (a new node's statistical information is used either from the beginning or after some information is gathered).
Experimental results on a representative set of halftone images, halftoned under different halftoning angles and at different resolutions, show that binary tree context modeling outperforms finite-context modeling by 5% to 10%. The gain in compression ratio achieved by using the adaptive template based on auto-correlation is of the order of 30% to 65% for binary tree context modeling.
Bibliography:
[1]The International Telegraph and Telephone Consultative Committee (CCITT),
Progressive Bi-Level Image Compression. Recommendation T.82, 1993.
[2] W. B. Pennebaker, J. L. Mitchell, G. G. Langdon, and R. B. Arps, "An
overview of the basic principles of the Q-coder adaptive binary arithmetic
coder," IBM J. Res. Devel., vol. 32, pp. 717--726, 1988.
[3] K. Denecker, S. Van Assche, and I. Lemahieu, "A fast autocorrelation
based context template selection scheme for lossless compression of halftone
images," in Proceedings of the Electronic Imaging Conference on Color
Imaging: Device Independent Color, Color Hardcopy, and Graphic Arts III,
1998.
[4] B. Martins and S. Forchhammer, "Bi-level image compression with
tree coding," in Proceedings of the IEEE Data Compression Conference
(J. Storer and M. Cohn, eds.), pp. 270--279, IEEE Computer Society Press,
Mar. 1996.
[5] K. Nguyen-Phi and H. Weinrichter, "Bi-level image compression using
adaptive tree model," in Proceedings of the IEEE Data Compression Conference
(J. Storer and M. Cohn, eds.), p. 458, IEEE} Computer Society Press, Mar.
1997.
Compressing CMYK images by removing color redundancy: A perceptual based quality evaluation
Author: ir. Peter De Neve
Phone: +32-9-264.89.08
Fax: +32-9-264.35.94
E-mail:
Extended abstract (Formulas and references are in LATEX style !!) One of the most common ways to represent color tones in continuous tone (contone) images is the CMYK color representation. The first three components are the subtractive complements of the additive primaries R, G and B. Because of several technical reasons a fourth component, black (K), is added. The addition of this fourth component introduces a higher tonal redundancy between the different color components.
Since 32 bits per pixel are used to represent the color information in these images and because they often have a high spatial resolution, compression might be considered. Because lossless compression schemes do not yield enough data reduction, lossy alternatives must be investigated.
When speaking of lossy compression, the current standard is JPEG \cite{Pennebaker93}. The only good way to compress CMYK color images with the current implementations is to process the four color planes independently from each other. In a three dimensional color space several algorithms have been proposed to decorrelate the color space, for instance the transformation from RGB to a luminance-chrominance YUV space. The mapping of the four dimensional CMYK color space to a three dimensional YUV space can't be used in our applications. The transformation is not reversible, since the information about one component is lost.
In \cite{cic.97} we have presented a technique that makes it possible to remove the tonal redundancy while staying in a four dimansional color space. In this technique the image is first divided into blocks since there will be a higher color correlation between pixels that are not too far from each other. The color decorrelation is performed block-wise by using a singular value decomposition (SVD) of the image data. From this approach we get a 4x4 orthonormal transformation matrix from which six rotation angles can be achieved. These can be quantized and have to be stored per block. The new color data can then be applied on a lossy compression technique.
The actual gain of this compression technique will be in the fact that due to the decorrelation of the image data, the most dominant color will be along the principal axis of the new color space. The other components perpendicular to that axis will be much smaller and can be coded with less bits.
In this paper we compare the standard JPEG technique, using optimized huffman tables, with the proposed technique in which the color space is first decorrelated before applying JPEG.
The distortion measures can be calculated by using the peak-signal-to-noise-ratio (PSNR), which is commonly used in lossy compression techniques. Nevertheless it has some disadvantages since it is widely accepted that the human perception is not "PSNR-like". Therefore the compressed algorithms are evaluated using more subjective quality criteria. We used the $\Delta$E error using a CIEL$^*$a$^*$b$^*$ metric \cite{CIE15.2}: $$ \Delta E = \sqrt{(\Delta L^*)^2 + (\Delta a^*)^2 + (\Delta b^*)^2} $$
We tested the algorithm on some well known test images. For the decorrelation of the color space we used a block-size of 32x32 \cite{cic.97}. In figure 1 the mean $\Delta$E (Left) and the maximum $\Delta$E (Right) are shown versus the compression ratio (CR) for the musicians image (1853x2103). In the quantization step of the compressor, we used the predefined quantization table available in JPEG. From both figures it can be seen that a significant gain is obtained by decorrelating the colors before performing JPEG that only reduces the spatial redundancy.
References (in Bibtex-style):
@book{Pennebaker93,
author = {Pennebaker, W.B. and Mitchell, J.L.},
title = {JPEG Still Image Data Compression Standard},
publisher = {Van Nostrand Reinhold},
year = {1993},
}
@TechReport{CIE15.2,
author = {CIE},
title = {Recommendations on uniform color spaces,
color-difference equations and psychonetric color terms},
institution = {Supplement No. 2 to Colorimetry, publication No. 15,
Paris: Bureau Central de la CIE},
year = {1978},
}
@InProceedings{cic.97,
author = {De Neve, P. and Philips, W. and Denecker, K. and Lemahieu, I.
},
title = {Introducing a decorrelated color space in the lossy compression
of pre-press applications},
booktitle = {The Fifth Color Imaging Conference: Color Science, Systems
and Applications},
year = 1997,
address = {Scottsdale, Arizona},
month = nov,
publisher = {IS\&T/SID},
pages = {88--91}
The Digital Graphic Network
Author: Phil Green
Phone: +44-171-5146500
Fax: +44-171-5146772
E-mail:
The Digital Graphic Network is an integrated collection of network-based services offered to the graphic arts and pre-press community by British Telecom. The specification of the services offered arose from a detailed independent analysis of the needs of the imaging community.
Printed documents almost invariably exist in digital form during their origination, and graphic arts businesses rely increasingly on telecommunications for the transfer of completed documents. There are, however, many other aspects of the business process between customers and suppliers that are suited to network-based interaction (Green, 1996; Groff 1997).
This analysis considered the features that a network-based transaction service would need to include in order to manage the transaction in an automated way, from initial client enquiry through to delivery of document originals into process queues, and receipt of payments into supplier accounts.
In this analysis, the interactions between client and supplier were classified into three streams: briefing, content and approval. The range of possible transactions at each stage were analyzed. The needs of the likely user community were reviewed in detail, and it was found that key requirements for the client include price competitiveness and assurance of capability on the part of the supplier; for the supplier they include credit risk and assurance that documents have been prepared correctly.
Without assurance that the producer and client will both deliver on their part of the contract, the transaction is seen as risky for both parties. If the entire business process transaction is to be transferred to the network, the responsibility for establishing producer and client capability have to be shouldered by the network. It is particularly important that completed documents being transferred into production queues are accompanied by comprehensive specifications and that approval processes include an element of pre-processing in order for production processes to be robust. It is also important to provide a means of updating and correcting documents.
Once a managed network service is in place, it is possible to extend it to include third-party interactions, such as download of objects such as high-resolution images, fonts, profiles etc from third-party content providers, directly to the output process queue as needed.
Many of the product service concepts proposed in this analysis are not yet implemented in the Digital Graphic Network and require further work in order to deliver them to the user community.
References Green, P. J. (1996) Professional Print Buying, GATF Press Groff, P. (1997) Understanding Electronic Communications: Printing in the Information Age, GATF Press
Using PDF in high-end colour reproduction
Authors: Agbasonu, A; Green, P. J; Hibbert, I; Linehan, S; Linford,
C; Sunderland,B.
Phone: +44-171-5146500
Fax: +44-171-5146772
E-mail:
The release of version 1.2 of the Portable Document Format in 1996 has led to expectations of its application in high-end colour reproduction. This is in addition to its established use as a cross-platform format for rendering low-resolution pages through new media such as the World Wide Web.High-end colour imaging is characterized by the combination ofhigh-resolution images and vector graphics in complex documents, with relatively high levels of accuracy in colour rendering. Devices on which such documents are output include imagesetters, platesetters, computer-to-plate systems and digital printers. These systems are currently running PostScript Level 2, and any changes in the imaging model should be capable of achieving the performance of the existing PostScript model as a minimum.
The PDF format has several important advantages over PostScript: it incorporates a degree of page independence absent in PostScript, making it easier to carry out page-level processing (for example imposing a document or splitting it up for processing over multiple RIPs); and the file format is more like a display list than a page description language stream, making the imaging model more robust and supporting soft proofing.
The revisions extend the file format to include support for typical high-end rendering requirements such as the inclusion of trapping, overprint and OPI information, and support for colour management. PDF now appears to have potential advantages in a high-end colour reproduction environment, but so far no systematic evaluation of its capabilities has been published. In this study, we analyze the relevant needs of colour imaging businesses and evaluate the ability of the PDF format to meet these needs. A test image is designed and output at a number of sites, including all current Level 3 RIPs and several Level 2 RIPs. The test image is output in both CMYK and RGB colour space to confirm the colour management capabilities.
The study analyzes the results of the tests, comments on the origin of the problems observed, and makes suggestions for future revisions in the format.
Print quality assessment by human perception
Author: Chan, J.
Phone: +44-171-5146500
Fax: +44-171-5146772
E-mail:
The objective of this project is to develop a method of simulating human print quality perception using an image analysis system.
Print quality measuring techniques have already been produced, but these only measure small parts of an image or a single variable at a time such as edge raggedness, noise or halftone frequency pattern. In order to make a better assessment of print quality perception as many variables as possible from the image should be considered simultaneously. For the last two years, a method has been under development to do this. This method uses an image analyser that consists of a CCD camera connected to a frame capture board and computer to collect data. This data is subsequently processed by neural network software.
The system analyses images of approximately 1cm2 in size. Initially, these images have been restricted to simple monochrome squares. The simple square has been chosen because it simplifies the programming that is necessary for the development of the data pre-processor stage of the system. The camera has been interfaced to the computer and a reliable lighting system for the image analyser has been developed. A data pre-processing algorithm has been developed in Visual Basic 5 for the MATLAB neural network software that will be used for the final data analysis.
Initial tests on the system show that it can determine changes in the quality of laser print when it is processed by a photocopier for a wide range of halftone patterns as well as solid print. The system can automatically measure halftone frequency patterns of laser printers operating at 600 dpi for their complete tonal ranges including solids. The system can also detect changes in the quality of laser printing by photocopying when other simple images such as circles, triangles and simple text characters are used.
The perceived changes in print quality in the tests described above have only used the author of this paper as an observer. Further work will include using other observers to test the system and a Turing test to validate the results.
Control Scheme For Printers Using More Than Three Color Inks
Author: Shoji Tominaga
Phone: +81-720-20-4562
Fax: +81-720-24-0014
E-mail:
The present paper describes a method for realizing an exact color reproduction on a printer using more than three color inks.
Background: Color reproduction requires color conversion between the color signals, depending on the device used, and the standard color coordinates like the CIE-L*a*b* color system, representing color appearance. Color reproduction on a CRT monitor is based on an additive mixture of three primaries. On the other hand, the color reproduction of print is based on the subtractive mixture of the three primaries Cyan ?, Magenta (M) and Yellow (Y), or four primaries with the addition of Black (K). It is usually difficult to predict from the knowledge of the amounts of ink for these primaries what color stimulus will be generated on paper. In other words, because of the complicated nonlinear relationship between the input and output signals of a printer, it is difficult to control the CMYK color signals in a simple operation using a matrix and a look-up table. Moreover recent color printers use many such color inks as CMYK plus light Cyan and light Magenta.
Method: A neural network method is presented for solving the mapping problem from the three-dimensional color space of color stimuli to the higher dimensional color space of printer signals. The CIE-L*a*b* color system is used as the device-independent color space. The mapping from the L*a*b* color space to the printer color space is constructed using a neural network. This mapping does not use such techniques as undercolor removal (UCR) and gray component replacement (GCR). The problem in four-color printing is considered as the problem of controlling an unknown printing system with four inputs and three outputs. A controller finds the CMYK signals necessary to produce the desired L*a*b* values from a printer. Our solution for this control problem is based on a two-phase procedure. The first phase determines a neural network for modeling the printer, and the second phase determines the combined NN system combining the printer model and the controller so as to provide the identity mapping. The controller then realizes the desired mapping for color control. Moreover, this technique is applied to the color control of a six-color printer using CMYK plus light Cyan and light Magenta.
Results:
Practical algorithms based on multilayer feedforward networks are presented. Firstly, a printer is modeled with a 4-10-10-3 type neural network having the four CMYK inputs and the three L*a*b* outputs. Secondly, the controller is constructed with a 3-10-10-4 type network to realize an inverse system of the printer model. The feasibility of the proposed algorithms are shown for a dye sublimation printer and an ink jet printer. The experimental results suggest that the controller achieves a color reproduction accuracy of less than 2.5 in the average L*a*b* color difference for both printers. Next, we use an ink jet printer which prints six colors of C, M, Y, K, Lt_C, and Lt_M. The color gamut of this printer is larger than that of the four-color printer. In this case, we have to determine the six input control color signals. It is shown that this control problem can be reduced to the four-dimensional color control. It realizes the accuracy of less than 2.0 in the average.
Discussion:
The six-color printing has a mapping problem from the three-dimensional color space of L*a*b* to the six-dimensional color space of printer signals C, M, Y, Y, Lt_C, and Lt_M. If we apply the original network method to this problem, then the printer is modeled by neural network with six inputs and three outputs, and the controller is constructed with a neural network with three inputs and six outputs. For example, the 6-10-10-3 type and 3-10-10-6 type networks might be used for the printer model and the controller network, respectively. However note that the learning procedure for determining the network parameters is complicated and time-consuming. We propose an effective method for solving this high-dimensional neural network problem. We analyze the surface-spectral reflectances of many color patches made by changing each value of the six color signals. Then it is found that the six-dimensional color control problem can be reduced to the four-dimensional problem, so that the same neural network algorithms as the ones for four-color printing are used for the present printer.
Biography: Shoji Tominaga was born in Hyogo Prefecture, Japan, on April 12, 1947. He received the B.E., M.S., and Ph.D. degrees in electrical engineering from Osaka University, Osaka, Japan, 1970, 1972, and 1975, respectively. Since 1976 he has been with Osaka Electro-Communication University, Neyagawa, Osaka, where he is currently a Professor with the Department of Engineering Informatics. In 1987-1988, he spent a year as a Visiting Scholar at the Department of Psychology, Stanford University. His current interests include computational vision and color image analysis. He is a member of IEEE, OSA, and ACM.
Document distribution and delivery across networks; digital watermarks, fingerprinting and traceability
Authors: Daniel Augot and Caroline Fontaine
Phone: +33 01 39 63 58 71
Fax: +33 01 39 63 53 30
E-mail:
Watermarking digital data appears to be a new and necessary way to protect intellectual property rights. Digimarc and Signum, two industry leaders in the domain, provide algorithms and softwares for embedding an invisible signal into images. Many teams are also working on the topic, and these technologies will appear as important as classical cryptology for protecting images, where usual cryptology fails. However, it appears that watermarking (or its aliases: fingerprinting, image authentication...) is a very difficult thing to do, since many operations which preserves appearance can be operated on an image, and it must be insured that the mark resists all these changes. This is not the issue we shall discuss here. We discuss here the problem of key management in a watermarking scheme. This discussion has been made in the Aquarelle European project, which we shall present briefly, and the state of these discussions is given here. It is a basic commandment in cryptology that an algorithm should be public, eventually parameterized by a key: the secrecy is in the key, not in the algorithm. We then distinguish different semantical aspects: watermarking, fingerprinting, monitoring and their significance on key signification and use. Next we focus on the simple process of watermarking, which is the process to add information relative to the copyright owner of the image. For that procedure, there is a key for embedding a signal into an image, which is also needed in order to recover the signal. For security considerations, we think it is best to have a different key for each image. We conclude that a Trusted Third Party is needed, which knows all the keys for incrustation. It has the role of verifying the presence of watermark in images, and is the only one to be able to do so.
At first, a simple functional model is introduced, in which copyright owners submit images to the Trusted Third Party. The Trusted Third Party watermarks images with a key, which is stored in a database. The watermarked image is then delivered back to the copyright owner. Anyone (most generally the copyright owner) can submit a suspected image to the Trusted Third Party, which retrieves the associated key and checks the mark. We show the major disadvantages of this scheme.
Next we consider a second and improved scheme, where the watermarking action is not done by the Trusted Third Party, but by the copyright owner itself. This induces that a key-exchange mechanism has to be chosen, and we have chosen the Diffie-Hellmann protocol. The whole protocol for key exchange and watermarking is name DHWM (Diffie-Hellmann and watermarking). We give some scenario to see this protocol at work.
Finally, we speculate on the functionalities (with respect to keys) an ideal watermarking algorithm should have. to enable an easy to copy with mechanism for the protection of images.
Colour Publishing in the Age of Digital Convergence
Author: Lindsay W. MacDonald
Phone: +44 1332 622222 x4085
Fax: +44 1332 622218
E-mail:
Considerable challenges lie ahead for colour publishing in the age of digital convergence. Over the next five years we shall see the integration of computing, television and communications into a single digital medium, in which enormous computing power will be combined with wireless connections, through terrestrial or satellite networks. The most natural way to fill the communications bandwidth will be with colour imagery, both still and moving.
Concomitant with the development of the channels will be the development of mobile information appliances. Within five years these will be readily available, at the price of today's laptop computers, and will provide full Internet access capability together with broadcast television and video-conference facilities, and telephone/voicemail services. Such appliances will be very attractive products for users and manufacturers alike, but will pose substantial challenges for both colour engineers and the publishers of content.
The issues for the colour imaging community range from image coding and compression through to the colour management across multiple media. For content providers and publishers the issues include devising new modes of interaction with and structuring of multimedia material (e.g. personalised information delivery services), through to watermarking and protection of image ownership. We are about to witness the most intense renewal of activity in image science and engineering since the birth of photography in the middle of the last century!
About the author: Lindsay W. MacDonald is Professor of Multimedia Imaging at the Colour & Imaging Institute at the University of Derby, UK, where he is responsible for the development of collaborative research and the planning of an MSc Colour Imaging course to commence in October 1998. Previously he was Professor of Multimedia Systems at the Cheltenham and Gloucester College of Higher Education. Until the end of 1995 he was for 18 years (1977-1995) with Crosfield Electronics Ltd. (now Fujifilm Electronic Imaging), the UK?s leading manufacturer of electronic pre-press systems for the professional graphic arts industry. In that role he designed and wrote the software for the world?s first computer-based page composition system (June 1977) and was a principal designer of the four generations of systems that followed, making significant contributions to their software architecture, image processing and human-computer interfaces. Professor MacDonald holds B.Sc. (Physics/Mathematics) and B.Eng. (Electrical) degrees from the University of Sydney and is a Fellow of the British Computer Society (FBCS), the Institution of Electrical Engineers (FIEE) and the Royal Society for the Arts (FRSA). In November 1997 he was General Co-Chair of the fifth annual IS&T/SID Color Imaging Conference in Scottsdale, Arizona. He is co-author or co-editor of four books. He also runs his own multimedia imaging consultancy
( NEW - will be presented in the Session 9 ) On the use of contextual Information in Production Processes
Authors: Michael Has, Christian Luidl, FOGRA, München,
Germany
The production process in graphic arts can be divided in two work- or dataflows which fundamentally differ. One deals with the creation and modification of content, i. e. photographing a scene, modifying text, retouch of picture data, etc. The other attempts to generate, organize, and manage contextual information which can e.g. be of technical (screening,color definition), administrative (customer, invoice info) or product specific information (desired paperstock, finishing) character. In the paper under view those workflows are closer examined. It is shown that the creation of genres (generic product prototypes like books, flyers etc.) may in certain cases be used to simplify both workflows. It may also be used to generate information required for both content creation (like layout) and technical purposes.
Last modified: 98/03/26 9:53:04