Development of a color management system for various visual environment

点击翻译按钮获取中文摘要

Color reproduction technologies have steadily been progressing over more than hundred years in the history of printing and photography. In recent years, a novel color reproduction technology, called color desktop publishing (DTP), has been developed. This technology permits reproduction of color images quickly and at a low cost. It utilizes personal computers and wide variety of color reproducing media such as printer or CRT monitor. However, the colors reproduced by each medium are strongly depend on the colorimetric characteristics of the colorants and transfer mechanism which vary from one meduim to another. It means that different printing devices produce different colors from the same device value, such as RGB or CMY(K), In addition to that, because the color appearance is changed under influence of viewing conditions, such as the white point and the luminance of the illuminant, the agreement of the color appearance, even as the same image, can no longer be obtained under differen … More t viewing conditions. This research describes a color conversion method for a transformation colorimetric to device value using a multi-layered neural network. It was shown that a neural network can accomplish a sufficient color conversion accuracy with relatively small number of data. To overcome illuminant-dependency problem, we adopt spectral reflectances for representing the colorimetric characteristics of the device. Then, we construct the multi-layered neural network which perform a color conversion from the device values to spectral reflectances, By using this neural network, we can achieve a flexible color conversion to the changing of the illuminant. This research also describes color gamut mapping technique as an optimization problem for finding an image which is perceptually closest to the original one among the images which only contain reproducible colors for the destination device. Perceptual difference between images is defined as a color difference of band-pass-filtered images rather than simple summation of color difference of each pixel within the image. Less