Showing papers presented at "Color Imaging Conference in 1996"

Learning Color Constancy

Abstract: We decided to test a surprisingly simple hypothesis; namely, that the relationship between an image of a scene and the chromaticity of scene illumination could be learned by a neural network. The thought was that if this relationship could be extracted by a neural network, then the trained network would be able to determine a scene's Illuminant from its image, which would then allow correction of the image colors to those relative to a standard illuminance, thereby providing color constancy. Using a database of surface reflectances and illuminants, along with the spectral sensitivity functions of our camera, we generated thousands of images of randomly selected illuminants lighting ‘scenes’ of 1 to 60 randomly selected reflectances. During the learning phase the network is provided the image data along with the chromaticity of its illuminant. After training, the network outputs (very quickly) the chro-maticity of the illumination given only the image data. We obtained surprisingly good estimates of the ambient illumination lighting from the network even when applied to scenes in our lab that were completely unrelated to the training data.

Analysis Multispectral Image Capture

Abstract: Multispectral image capture (i.e, more than three channels) facilitates both more accurate tristimulus estimation and possibilities for spectral reconstruction of each scene pixel. A seven-channel camera was assembled using approximately 50 nm bandwidth interference filters, manufactured by Melles Griot, in conjunction with a Kodak Professional DCS 200m digital camera. Multichannel images were recorded for the Macbeth ColorChecker chart as an illustrative example. Three methods of spectral reconstruction were evaluated: spline interpolation, modified-discrete-sine-transformation (MDST) interpolation, and an approach based on principal-component analysis (PCA). The spectral reconstruction accuracy was quantified both spectrally and by computing CIELAB coordinates for a single illuminant and observer. The PCA-based technique resulted in the best estimated spectral-reflectance-factor functions. These results were compared with a least-squares colorimetric model that does not include the spectral-reconstruction step. This direct mapping resulted in similar colorimetric performance to the PCA method. The multispectral camera had marked improvement compared with traditional three-channel devices. Techniques The use of multispectral cameras can lead to dramatically improved colorimetric accuracy in comparison with three-channel devices. 1,2

Retinex Image Processing: Improved Fidelity to Direct Visual Observation

Abstract: Recorded color images differ from direct human viewing by the lack of dynamic range compression and color constancy. Research is summarized which develops the center/surround retinex concept originated by Edwin Land through a single-scale design to a multi-scale design with color restoration (MSRCR). The MSRCR synthesizes dynamic range compression, color constancy, and color rendition and, thereby, approaches fidelity to direct observation.

The Maximum ignorance assumption with positivity

Abstract: Color devices are not colorimetric. It follows then that some color correction must be done to map device RGBs to XYZ’s. One common correction method involves finding the linear transform which takes device spectral sensitivities as close to the XYZ color matching curves as possible (in the least-squares sense). Thereafter, this transform is used to map device RGBs to XYZs. It is well known that this procedure is statistically justified so long as one assumes that all spectra with positive and negative power are equally likely to occur i.e., so long as one is maximally ignorant about the world. In this paper we point out that the maximally ignorant stance is unjustified since spectra with negative power cannot physically occur. This leads us to develop the notion of maximal ignorance with positivity i.e., we assume that all spectra which are everywhere all positive are equally likely. We demonstrate that this new maximal ignorance stance delivers considerable benefits in terms of improved color correction.

Is Color Constancy Task Independent

Abstract: The goal of color appearance models is to predict how observers perceive, describe, and match colors. Two difficulties prevent the easy formulation of these models. First, the color appearance of a light can depend strikingly on the context in which it is viewed. Thus color appearance models must incorporate transformations to account for context effects. Second, there are multiple ways that observers can assess and report color appearance. A complete model must predict performance for all appearance tasks. Color appearance models can be simplified if the effects of context may be separated from how appearance is measured. One way to achieve such separation is to formulate the model in two distinct stages. The first stage incorporates context effects by transforming tristimulus coordinates to appearance coordinates. This transformation is allowed to depend on viewing context but not on how appearance is measured. The second stage maps between the appearance coordinates and actual observer responses. Different mappings are allowed for different tasks, but each mapping is restricted to be context independent. The RLAB color appearance model is consistent with this type of separability. Separating the effect of context from task specific mappings allows model parameters to be determined economically. The effects of context need only be measured using one task, while the form of the task dependent mappings need only be characterized for one context. Of course, it only makes sense to incorporate such context-task separability only if models of this form actually describe performance. In this paper we report initial empirical tests. Introspection suggests that context-task separability is likely to hold. As we experience the world we are generally aware of a unitary color percept at each image location. It is appealing to think that this subjective impression mediates our performance on all color tasks. There are several reasons why we feel an empirical check is desirable. First, although introspection is useful it is important to build our models on the firmer foundations of experimental data. Second, there are conditions under which cognitive factors such as instructions or perceptual set can influence color judgments. The influence of cognitive factors may be both context and task dependent and their influence could provide a mechanism through which separability could fail. Third, Troost and DeWeert have presented data which, while not conclusive, call the separability hypothesis into doubt. In our experiments, we use three tasks to measure the effect of changing the illuminant on color appearance. In the first task, observers adjusted a surface to appear achromatic. In the second task, observers performed asymmetric matches. In the third task, observers named colors. We used the same observers and stimulus conditions for all three tasks. We then asked whether the effect of the illuminant was independent of task.

Effects of Ambient Illumination on the Appearance of CRT Colors.

Abstract: We investigated the effects of ambient illumination on the color appearance of CRT which was positioned in a room illuminated by experimental illuminants, F2, F5, and F8. The luminance of these illuminants were controlled at three levels. The observers adjusted a color patch on the CRT to find achromatic matching on a random color background. The chromaticity of the achromatic color was shifted toward that of the illuminant, which implies the change of the observer's chromatic adaptation state. The shift ratios, calculated in the uniform space u'-v', increased according to the increase of the distance between chromaticities of CRT white and those of illuminants and increased according to the increase of luminance level of the illuminants. These results were implemented in a monitor profile for a color management system to get more consistent color appearance of the CRT which was viewed under various ambient illuminations.

Why is Black-and-White so Important in Color?

Abstract: In the visual system, the retina communicates with the brain by means of a black-white (achromatic) signal and two color-difference signals, a red-green and a yellow-blue. The existence of the black-white signal has important implications in imaging. First, if areas that are intended to be black. gray, or white are reproduced with even a slight tinge of hue, the defect is usually very noticeable. because these achromatic perceptions correspond to the color-difference signals being balanced at their null levels. Second the achromatic signal largely determines the apparent contrast of scenes, and their images only look correct if their gray scales are adjusted with due allowance for the effect of the surround on the black-white signal. Third, in luminancechrominance television, important reductions in bandwidth are possible because of the lower sharpness required in the chrominance signals as a consequence of the greater number of cones necessary to generate the color-difference signals than to generate the black-white signal; the extent to which advantage is taken of this situation is discussed in connection with various forms of imaging.

Psychophysical Generation of Matching Images for Cross-Media Color Reproduction.

Abstract: A technique was developed and tested for generating CRT pictorial image reproductions that match the color appearance of print originals. Five observers, experienced in using Adobe Photoshop to adjust images, matched two scenes using a memory-matching technique Observers were able to produce accurate matches when originals and reproductions were viewed at the same white point. Observers then matched the reproductions at 6500 K to originals viewed at 9300 and 3000 K. These matches were compared to the predictions of various color-appearance models, using a paired-comparison technique in a psychophysical experiment using different observers. The observer-matched images were found to be equal or superior to matches produced using any of the models.

Color Management for Color Facsimile.

Abstract: We propose a color management system for the color facsimile. It consists of protocols for the colorimetric calibration of the scanner and the printer by establishing the relationships between the device-dependent color coordinates and the device-independent CIELAB color space. The scanner calibration is based on 3rd order polynomial regression techniques. The printer calibration uses 3D triangulation techniques, which gives us high flexibility and which allows us to apply different color gamut mapping techniques in an efficient way. The online color transformations are calculated by a tetrahedral interpolator using 3D look-up tables provided by the calibration algorithms. Introduction The first international standard for color facsimile was approved in 1994. It fixed JPEG as the encoding method for continuous-tone color images and chose the CIELAB color space with CIE Standard Illuminant D50 as the default color space for data transmission. An important aspect of color facsimile products is the quality of the color reproduction. To obtain faithful color reproduction, colorimetric calibration of the I/O devices is needed, so that the device-dependent color coordinates of the scanner, the printer and the monitor can be linked to the device-independent CIELAB color space. Furthermore, efficient means for processing and converting images are needed. These are typically the tasks performed by a color management system. To calibrate I/O devices colorimetrically, three different approaches are typically used: • Theoretical/physical models. Classical examples of this approach are the Neugebauer printer model, and the monitor models based on relatively few parameters such as the specification of the chromaticity coordinates and gamma value of each channel, and the specification of the white point. The drawback with such models is that they are mostly technology dependent. • Analytical models. They are based on the minimizing of the mean square error of a set of measurement points by polynomial regression. These models are widely used, especially for input devices such as scanners. Important design parameters for this approach are the polynomial order as well as the choice of input/output color spaces requiring pre/post-processing. • Geometric models. Such models are based on lookup-tables and interpolations. They are used by numerous authors, and have the advantage to be general. However, they require mostly a larger amount of measured data, compared to the precedent models. We present a color management system for the color facsimile, consisting of procedures for the colorimetric calibration of the scanner, based on an analytical model, as well as for the calibration of the printer, based on a geometric model. The calibration of the monitor, based on a simple physical model of a CRT display is quite straightforward, and will not be presented in this document.

Color Halftoning with Blue Noise Masks.

Abstract: Color halftoning using a conventional screen requires tating the screen by different angles for different co planes to avoid Moiré patterns. An obvious advantag halftoning using a Blue Noise Mask (BNM) is that the are no screen angles or Moiré patterns. However, a si strategy of employing the same BNM on all color pla is unacceptable in cases where a small registration can cause objectionable color shifts. In a previous pa we proposed shifting or inverting the BNM for differe color planes. The shifting technique can, at certain s values, introduce low frequency contents into the h tone image, whereas the inverting technique can be only on two color planes. In this paper, we propose a t nique that uses four distinct BNMs that are correlated way such that the low frequency noise resulting from interaction between the BNMs is significantly reduce

Algorithm-Independent Color Calibration for Digital Halftoning.

Abstract: A novel method based on measuring 2 × pixel patterns provides halftone-algorithm independent color calib tion for digital halftoning. The binary CMY(K) colo signals can be mapped into CIE XYZ color space at printer resolution level. Therefore, any binary CMY( color images can be described as continuous-tone ages in standard color spaces. The new method has successfully applied to halftone screen calibration, v tor error diffusions, as well as stochastic screens.

CIE Colorimetry and Colour Displays.

Abstract: CIE colorimetry was standardised 65 years ago to describe colorimetric properties of signal lights. The first real application of CIE colorimetry occurred, however, in the textile and coating industries to describe colour matches between sample and reference. Thus also the experiments conducted to achieve a uniform colour scale diagram were performed by using material samples usually illuminated by a daylight simulator. Such experiments led to the recommendations for the CIELAB and CIELUV colour spaces. It has been a misinterpretation that CIELUV is more useful for self-luminous objects, derived from the fact that CIELUV has a chromaticity diagram and CIELAB has non. In recent years much effort has been invested to obtain a colour appearance model to be able to describe not only colour matches but also the appearance of colours, both as material samples and in order to be reproduced on colour displays. The three fundamental questions in this respect are: • Are the standard colour matching functions (cmf’s) representative of average human colour vision? • What is the proper description of chromatic adaptation? • What are the auspices of a CIE colour appearance model? A number of CIE Technical Committees are working on the different aspects of these questions. The paper presents a short review of the questions raised and, as far as feasible, will outline likely answers.

Predicting the spectral behaviour of colour printers for transparent inks on transparent support

Abstract: A spectral colour prediction model is proposed and applied to an ink jet printer. We limit the field of investigation to transparent inks printed on a transparent substrate. The proposed method is based on the computation of the transmittance spectra of colour primaries applying Beer’s law, and on the calculation of their respective surface coverage using a high resolution grid. The proposed prediction model requires measuring the transmittance spectra of the inks and approximating the density distribution function of a single printed dot of each ink. We hereby obtain accurate spectral predictions of colour patches, while the number of required sample measurements is reduced to a minimum.

The Color Gamut of Halftone Reproduction.

Abstract: Color mixing by a halftoning process, as used for color reproduction in graphic arts and most forms of digital hardcopy, is neither additive nor subtractive. Halftone color reproduction with a given set of primary colors is heavily influenced not only by the colorimetric properties of the full-tone primaries, but also by effects such as optical and physical dot gain and the halftone geometry. We demonstrate that such effects not only distort the transfer characteristics of the process, but also have an impact on the size of the color gamut. In particular, a large dot gain, which is commonly regarded as an unwanted distortion, expands the color gamut quite considerably. We also present an image processing model that can describe and quantify the effects of physical and optical dot gain on different media and with different halftoning methods.

Hole removal in high-pH-induced reduction-sensitized silver bromide grains

Abstract: High-pH-induced reduction sensitization was used to form silver clusters on the surfaces of octahedral silver bromide grains. This method had the advantage of not leaving unreacted reducing agent in the emulsion, which could cause further silver cluster formation over time. Diffuse reflection spectroscopy measurements showed these silver clusters to be similar to those formed by dimethylamine borane and SnCl 2 , except that the absorption peaks were noticeably less intense. Large emulsion grains sensitized in this manner benefited most from the hole-removal ability of the silver clusters. The result was a linear relationship between speed and grain volume for edge lengths 0.28 to 1.22 μm, with a slope of 0.93 +/- 0.08, which is close to the theoretical slope of 1.0. Core-shell emulsions in which the cores were oversensitized with sulfur also benefited from the high-pH-induced reduction sensitization on the surface. Instead of desensitization of the core when oversensitized, these emulsions showed dramatically increased internal sensitivity, consistent with the proposed hole-removal property of the surface silver clusters. No evidence of electron trapping by these chemically produced silver clusters could be found.