In this work, we investigate the use of a digital camera for colorimetry. Our system consists of a measurement device and a corresponding calibration mapping. The goal is to design a system that will accurately assess the color of a sample. We develop two colorimetry systems by applying model-based and regression-based techniques. For both systems, the measurement device is formed by a digital camera and a set of filters. The term multi-exposure refers to the multiple snapshots taken by the camera along with filters. The calibration mapping which consists of matrices then takes these filtered camera RGB outputs, and returns the CIE XYZ tristimulus values under several pre-selected illumination conditions. For the model-based system, a model for the measurement device is employed; and our objective is to find the optimal filters and the corresponding calibration sets that minimize a cost function which accounts for errors in L*a*b* space, system robustness, and filter smoothness. For the regression-based system, no modeling technique is applied to the measurement device. The objective is simply to find the optimal calibration matrices that minimize the total least squared errors of a given color set in CIE XYZ coordinates under several pre-selected illumination conditions. We apply both types of colorimetry systems to two specific tasks: general purpose measurement of color samples and colorimetry of human teeth. We present experimental results for both applications. Finally, in order to measure the parameters for these systems and evaluate their performance, we had to develop special instrumentation. We will briefly describe this effort as well.

Principal component analysis coupled with multi-channel digital image capture is a powerful technique for spectral scene estimation. Most often, the linear modeling is performed using a spectral reflectance factor. However, it is well known that for many subtractive coloration systems, spectral reflectance is nonlinearly related to colorant amount. Accordingly, the accuracy of spectral reconstruction has been evaluated as a function of the spectral definition of the ensemble. Specifically, Kubelka-Munk turbid media theory and a new empirical transformation, optimized for optimal data normality, were compared with spectral reflectance factor. Both tested spaces are nonlinear transformations of spectral reflectance factor. In addition, a new technique of multi-channel digital image capture was developed and tested. This technique combined trichromatic image capture with color filtration resulting in multiple signals in sets of three. Six eigenvectors based on the new empirical space coupled with digital capture with and without a light-blue absorption filter produced the most accurate spectral scene estimation from among the various tested combinations.

Color images often must be color balanced to remove unwanted color casts. Color balancing uncalibrated images (e.g. downloaded from the Internet or scanned from an unknown film) adds additional challenges to the already difficult problem of color correction because neither the pre-processing to which the image was subjected, nor the camera sensors or camera balance are known. In this article, we propose a framework for dealing with some aspects of this type of image. In particular, we discuss the issue of color correcting images where an unknown ‘gamma’ non-linearity may be present. We show that the diagonal model, used for color correcting linear images, also works in the case of gamma corrected images. We also discuss the influence that unknown camera balance and unknown sensors have on color constancy algorithms. To perform color correction on uncalibrated images, we extend previous work on using a neural network for illumination, or white-point, estimation from the case of calibrated images to that of uncalibrated images of unknown origin. The results show that the chromaticity of the ambient illumination in uncalibrated images can be estimated with an average CIE Lab error around 5ΔE. Comparisons are made to the grayworld and white-patch methods.

At an early stage in almost all color reproduction pipelines, device RGBs are transformed to CIE XYZs. This transformation is called color correction. Because the XYZ color matching functions are not a linear combination of device spectral sensitivities there are some colors which look the same to a device but have quite different XYZ tristimuli. That such device metamerism exists is well known, yet the problem has not been adequately addressed in the color correction literature. In this article, we examine in detail the role that metamers play in developing a new color correction algorithm. Our approach works in two stages. First, for a given RGB we characterize these possible camera metamers. In the second stage this set is projected onto the XYZ color matching functions. This results in a set of XYZs any one of which might be the correct answer for color correction. Good color correction results by choosing the middle of the set. We call the process of computing the set of metamers, projecting them to XYZs and performing the selection, metamer constrained color correction. Experiments demonstrate that our new method significantly outperforms traditional linear correction methods. For the particular case of saturated colors (these are among the most difficult to deal with) the error is halved, on average, and the maximum reduced by a factor of four.

The lack of a scanner calibration target representative of the image medium (substrate and colorants) often limits the accuracy of scanner color calibration. In this paper, a color calibration method is presented for photographic input media that does not require a calibration target. Using characteristic spectral measurements from the image(s) to be scanned, a model for the spectra on the medium is obtained through a principal omponent analysis. The spectral sensitivity of the scanner provides a model for its operation. By combining the models for the scanner and the medium, the spectral reflectance of the input corresponding to a given set of scanner RGB values can be determined. This provides a spectral calibration for the scanner, which can readily be transformed into a color calibration under any suitable viewing illuminant. Results from simulations and actual calibrations demonstrate the value of the new method.

This article proposes an illuminant estimation algorithm that estimates the spectral power distribution of an incident light source from a single image. The proposed illumination recovery procedure has two phases. First, the surface spectral reflectances are recovered. In this case, the surface spectral reflectances recovered are limited to the maximum achromatic region (MAR) which is the most achromatic and highly bright region of an image, after applying intermediate color constancy process using a modified gray-world algorithm. Next, the surface reflectances of the maximum achromatic region are estimated using the principal component analysis method along with a set of given Munsell samples. Second, the spectral distribution of reflected lights of MAR is selected from the spectral database. That is, a color difference is compared between the reflected lights of the MAR and the spectral database, which is the set of reflected lights built by the given Munsell samples and a set of illuminants. Then the closest colors from the spectral database are selected. Finally, the illuminant of an image can be calculated dividing the average spectral distributions of reflected lights of MAR by the average surface reflectances of the MAR. In order to evaluate the proposed algorithm, experiments with artificial and real captured color-biased scenes were performed and numerical comparison examined. The proposed method was effective in estimating the spectral distribution of the given illuminants under various illuminants and scenes without white points.

Color transformations in digital imaging systems are often implemented with lookup tables (LUTs) that require some form of multidimensional interpolation. Such LUT based transformations typically involve a trade-off between the computational cost, required storage and/or memory, and the resulting accuracy of the transform. In this article, novel methods are proposed for improving some of these quality-cost trade-offs. The methods fall in two categories: (i) those that improve the trade-off between computational cost and quality; and (ii) those that enhance the trade-off between LUT size and quality. Results show that promising trade-offs can be achieved by exploiting the properties of the human visual system, as well as the characteristics of the function being approximated by the LUT.

Gamut mapping is a technique to transform out-of-gamut colors to the inside of the output device's gamut. It is essential to develop effective mapping algorithms to realize WYSIWYG (What You See Is What You Get) color reproduction.¹ We had previously found that three-dimensional gamut mapping is superior to the two-dimensional mapping, when we applied Mahalanobis distance as a color difference equation, such as BFD color difference formula.^2,3 However, in our previous experiments, a clipping method was used that maps all out-of-gamut colors to the surface of the gamut, and no change was made to colors inside the gamut. Since this method could possibly cause loss of gradation in an image, we had investigated non-linear compression for the three-dimensional gamut mapping in this study. The results of visual experiments indicated that preferred compression method depends on image contents. If the saturated colors that are out-of-gamut contain high frequency components, a certain degree of compression was needed. On the other hand, if those colors only have gradual change with low frequency components, clipping method was preferred.

This article describes an investigation of several chroma compression algorithms with constant and modified lightness and hue. While the rescaling of differing lightness ranges is not considered here, gamut-mapping algorithms with different mapping types and mapping directions are evaluated. Among these are the methods of mapping colors towards a focal point as well as the newly developed relative lightness change (RLC) technique. The latter maps colors along curved lines and enables the direct use of cylindrical coordinates lightness, chroma, and hue for the mapping, regardless of the mapping direction. Moreover, the RLC method is straightforward and does not need any iteration or complex cross sectioning, thus it is well suited for real time gamut mapping. Issues of gamut mapping such as the mapping space, the coordinate system within the space, and gamut boundary description are discussed. Psychophysical experiments are described which were conducted to evaluate the reproduction of monitor images on a device with a clearly smaller color gamut. All images were viewed on the monitor. The algorithms were dependent on the image gamuts. The experiments showed that the RLC method performs superior compared to the focal mapping algorithms. Moreover, the optimum mapping direction was RLC50 that means a slight adaptation of lightness. The best mapping type was pure clipping, regardless of the mapping direction. Notably, RLC50 outperformed other, recently published algorithms. While the found algorithms may still require some fine-tuning, a thorough understanding of the mechanisms of gamut mapping was acquired.