Of the three basic color perceptions, hue, brightness, and colorfulness, hue has no relative version, but brightness has lightness, and colorfulness has chroma and saturation. Correlates of chroma are widely used in color difference formulae, but saturation currently plays little part in color science and technology. This is perhaps because in many industries flat samples are viewed in uniform lighting for the evaluation of color differences, and in this case chroma is the appropriate contributor for samples of small angular subtense. For samples of large angular subtense, however, a correlate of saturation may be more appropriate to use. In the real world, it is common for solid objects to be seen in directional lighting; in these circumstances saturation is a more useful percept than chroma because the former remains constant in shadows. In imaging, artists and computer-graphics operators make extensive use of series of colors of constant saturation. In optical imaging, saturation can be an important percept in large dark areas. Recent experimental work has provided a much improved correlate of saturation.

Using color names is one of the most important tools to communicate color information. We carried out a color naming experiment, where subjects were asked to sort 292 Munsell color samples under various light sources into eleven basic colors defined by Berlin and Kay. The experimental results were represented in various color spaces, the Munsell color space, the CIE (x, y) chromaticity diagram, the CIELAB space and the CIECAM97s (J, C, h) space. Performance of these color spaces for representing color names will be discussed.

The ultimate in color reproduction is a display that can produce arbitrary spectral content over a 300-800 nm range with 1 arc-minute resolution in a full spherical hologram. Although such displays will not be available until next year, we already have the means to calculate this information using physically-based rendering. We would therefore like to know: how may we represent the results of our calculation in a device-independent way, and how do we map this information onto the displays we currently own? In this paper, we give an example of how to calculate full spectral radiance at a point and convert it to a reasonably correct display color. We contrast this with the way computer graphics is usually done, and show where reproduction errors creep in. We then go on to explain reasonable short-cuts that save time and storage space without sacrificing accuracy, such as illuminant discounting and human gamut color encodings. Finally, we demonstrate a simple and efficient tone-mapping technique that matches display visibility to the original scene.

A psychophysical experiment was performed to determine the psychological dimensions involved in judging image quality. Seven different prints for each of two images, a portrait and a landscape, were produced using a combination of 5 printers and different paper types. The experiment consisted of two parts that were run concurrently. In the first part, paired-comparison was used to evaluate image preference. In the second part, judgments of similarity and dissimilarity were made using triad presentations. The paired-comparison data were analyzed using Thurstone's Law of Comparative Judgment and Dual Scaling, a multidimensional statistical technique that reveals the independent dimensions used in categorical judgments. The judgments of similarity and dissimilarity were analyzed using nonmetric multidimensional scaling. The results indicate that the psychological stimulus space can be characterized well in two dimensions. An ideal point model can be used to identify preference in this space. Variation in subjects' preferences can be characterized predominantly in one dimension and the subjects are fairly consistent in their response along this dimension. The psychological stimulus space correlated highly with color variation in the images. We conclude that multidimensional techniques can be used to analyze image preference and find relationships between psychological and physical variables relating to image quality. Specifically, our results indicate that color is of primary importance for judging image quality in our particular situation.

Illuminant estimation in natural scenes includes the problem of estimating a spatial distribution of light sources by omnidirectional observations. The present paper describes a method for estimating an omnidirectional light distribution from the image of a camera aiming at a mirrored ball. The parameters including the camera location in the scene and the focal length of the camera lens are estimated to precisely determine the mapping between coordinates on the ball and light rays in the world. We create two types of image to represent the omnidirectional light distribution in the world. One representation is a polar coordinate system with the origin located at the center of the mirrored ball. The other is a parallel projection on a two-dimensional screen passing the center of the ball. Experiments in a natural scene using a spherical steel ball and a color CCD camera are described. First, we estimate an omnidirectional radiance distribution indoors in a room with fluorescent ceiling lights. Second, we estimate the radiance distribution in the open air. The scene includes strong specular high light by direct reflection of sunlight. We analyze the chromaticity of the omnidirectional light source.

We introduce an active imaging method to measure scene illumination. The system implementation is divided into four steps. First, the system acquires two images: one is an ordinary image of the scene under ambient light and the other is a corresponding image in which light from the camera flash is added to the scene. Second, the image pair is analyzed to obtain an image that represents the scene as if it had been illuminated by the flash alone. Third, the flash-only image is used to estimate object reflectance functions. Fourth, using the estimated reflectance functions, the ambient illumination spectral power distribution is estimated. We present results that evaluate the method's stability with respect to changes in the mean reflectance function of the scene. Finally, we discuss limitations of the current implementation and alternative implementations.

Annotating images with a description of the content can be useful in processing images, by taking into account the scene depicted. We show here that it is possible to relate low-level visual features to semantic photo categories, such as indoor, outdoor and close-up, using CART classifiers. We have designed and experimentally compared several classification strategies, producing a classifier that can provide a reasonably good performance and on generic photographs no matter how acquired.

In this paper, we present a novel concept of color correction for consumer digital still camera (DSC) images. This concept is based on a hierarchical Bayesian image content analysis consisting of feature extraction and unsupervised clustering and on a set of color correction algorithms that have been optimized on the obtained characteristic image classes. Since the concept uses Bayesian inference to combine several color correction results, further available information (e.g., obtained from camera metadata) can be easily integrated into the color correction process.

The physical properties of color are usually described by their spectra, eigenvector expansions or low-dimensional descriptors such as RGB or CIE-Lab. In the first part of the paper we show that many of the traditional methods can be unified in a framework where color spectra are elements of an infinite-dimensional Hilbert space that are described by projections onto low-dimensional spaces. We derive some fundamental geometrical properties of the subset of the Hilbert space formed by all color spectra. We describe the projection operators that map the elements of the Hilbert space to elements in a finite dimensional vector space. This leads to a generalization of the concepts of spectral locus and purple line. It will be shown that for geometrical reasons the color space is topologically equivalent to a cone. In the second part of the paper we illustrate the theoretical concepts with four large databases of spectra from color systems and a series of multi-spectral images of natural scenes. We verify the conical property of color space for these databases and compute their, geometrically defined, spectral locus and chromaticity properties. In the last section we relate the natural co-ordinate system in the conical color space to the traditional polar co-ordinates in CIELab. We show that there is a good agreement between the geometrically defined hue-variable and the angular part of the polar co-ordinate system in CIE-Lab. There is also a clear correlation between the geometrical and the CIE-Lab saturation descriptors.

The present paper describes an experimental apparatus and set of calculations for determining various parameters of the surface reflection model. The apparatus consists of a lighting system to emit parallel beams, a goniometric rotating table, and a vision system with six spectral channels. The camera data are used to estimate various parameters such as (1) spectral reflectance, (2) illuminant spectrum, (3) surface roughness, (4) ratio of body to interface intensity, and (5) index of refraction. First, a method for estimating (1)-(4) from a single multband image is introduced. Second, the parameter (5) is estimated using highlight peaks of an object surface at different conditions of illumination and viewing. Experimental results are shown using the curved surface of a painted object. The overall feasibility of our method is confirmed on computer graphics images created by using the estimated parameters.