Development of a color management system for various visual environment

在一百多年的印刷和摄影历史上，彩色再现技术一直在稳步发展。近年来，一种新的色彩再现技术——彩色桌面印刷（DTP）得到了发展。这种技术可以快速、低成本地再现彩色图像。它利用个人电脑和各种各样的彩色再现媒体，如打印机或CRT显示器。然而，每种介质所再现的颜色在很大程度上取决于着色剂的比色特性和从一种介质到另一种介质的传递机制。它是指不同的印刷设备从相同的设备值（如RGB或CMY(K)）产生不同的颜色，此外，由于颜色外观在观看条件（如白点和光源亮度）的影响下发生了变化，因此即使是同一图像，在不同的观看条件下也无法获得颜色外观的一致性。本研究描述了一种利用多层神经网络将比色计转换为器件值的颜色转换方法。结果表明，神经网络可以在相对较少的数据量下实现足够的颜色转换精度。为了克服光源依赖问题，我们采用光谱反射率来表示器件的比色特性。然后，我们构建了多层神经网络，将器件值转换为光谱反射率，利用该神经网络可以根据光源的变化实现灵活的颜色转换。本研究还将色域映射技术描述为在仅包含目标设备可复制颜色的图像中寻找感知上最接近原始图像的优化问题。图像之间的感知差异被定义为带通滤波图像的色差，而不是图像内各像素色差的简单相加。少

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