A survey is given of the paths by which additive and subtractive color reproduction in photography, television, and printing, have led from their early beginnings to their present states. The outlook for the future in each field is reviewed.

The human retina encodes information about images through the responses of three classes of photoreceptor, often referred to as the L, M, and S cones. These photoreceptors are arranged in three interleaved mosaics; at any one retinal location only a single cone type samples the retinal image. To create our percept of a continuous colored world, the visual system must reconstruct the responses of the missing two cone types at each retinal location. The algorithm that performs this reconstruction works very well—we rarely perceive artifacts that arise from the interleaved sampling arrangement.Most CCD based color cameras employ the same interleaved sampling architecture as the human retina. Yet for CCD cameras, color artifacts are quite common near sharp luminance boundaries. CCD cameras are more susceptible to these artifacts because their reconstruction algorithms are not as successful as the one employed by the human visual system.This talk will begin by reviewing basic research designed to elucidate reconstruction by the human visual system. We will then show how ideas that emerged from the basic research have led to a new algorithm for processing images acquired with CCD cameras.

Induction refers to the change of the color appearance of a light caused by the presence of neighboring lights. There are two major types of induction: color contrast, which occurs when the color appearance of a light shifts away from the color of the neighboring lights; and color assimilation, which occurs when the color appearance of a light shifts toward the color of the neighboring lights.Previous studies reported that spatial properties are the main factor determining the transition from assimilation to contrast. With high spatial frequency stimuli, assimilation occurs. With low spatial frequency stimuli contrast occurs. With intermediate spatial frequencies, variable effects are seen. The purpose of this study is to investigate the mechanisms of the induction. To achieve this goal we measured the induction effects for different stimulus spatial frequencies (0.8, 4.0, 6.0, and 9.0 cpd) to determine the transitional spatial frequency between assimilation and contrast. The use of a cone excitation space allows analysis of the spatial frequency effects on SWS and LWS/MWS cones chromatic pathways separately. A spread light model was developed to see whether optical factors might account for all or part of the assimilation effect.

We present a non-recursive integral-equation model that predicts perceived grey levels in complex achromatic displays as a function of the physical luminances of individual pixels in the display. The model incorporates spatially local luminance adaptation mechanisms, contrast gain controls and spatial distance dependent weights on lateral connections, and linear summation of the induced effects of individual surrounding pixels.

A visual experiment was completed to evaluate the degree of observer metamerism in color matches between typical color reproduction media. The results illustrate that current CIE standard colorimetric observers provide reasonable estimates of average color matches and that the range of color mismatches encountered in cross media color reproduction due to observer variability can be very large, on the order of 10 CIELAB units.

Human visual system is partially adapted to the CRT monitor's white point and partially to the ambient light, when comparing a soft-copy image with the reproduced hard-copy image. The visual experiments were performed on the effect of the ambient lighting under mixed chromatic adaptation. It was found that human visual system is 40% to 60% adapted to CRT monitor's white point and the rest to the ambient light. This adaptation ratio itself was found to be independent of the luminance level of the ambient lighting for the portrait images.Practical method for appearance match between soft-copy and hard-copy is presented in this paper. This method is fundamentally based on a simple von Kries' adaptation model and in addition, takes into account of the human visual system's “partial adaptation.”

Linear models appear in many recent color constancy theories; however, they can play two quite different roles. They may either occur as a direct component of the computational strategy, as in the case of the Maloney-Wandell algorithm; or they may appear as necessary part of the theoretical development, but not as part of the computation perse, as in the case of algorithms based on spectral sharpening (Finlayson, Drew, Funt). This paper surveys recent work in color constancy and concludes that in many circumstances the second role may be more appropriate than the first.

We suggest dichromatic reflection models adequate for describing surface-spectral reflectances of a variety of materials. First we describe the standard dichromatic reflection model for inhomogeneous dielectric materials. Next the generalized models are shown for describing the spectral reflectances of the materials for which the standard model is inadequate. The reflection characteristics are analyzed on a chromaticity diagram. We demonstrate practical use of the reflectance models in computer graphics and color constancy.

Local color pixel distributions provide information that is useful for object recognition but are dependent on the scene illumination. We develop a method that assigns color descriptors to an object that depend on the distribution of spectral reflectance across the object and not on the illumination. For a trichromatic system, the method assumes a three-dimensional linear model for surface spectral reflectance. We present examples demonstrating the system's ability to recognize model objects in cluttered scenes independent of scene illumination.

For exact color reproduction of objects under varying illuminants it is necessary to provide information on the complete reflectance spectrum. Since no analytical solution exists a stochastic optimization algorithm is applied to calculate basis vectors for representing a set of test spectra that lead to a minimal color deviation.