A Cluster-Based Framework for Contrast Preservation During Color Space Transformations

Common tasks such as printing, displaying, and visualizing images may require a transformation of the input image from its source color space to a target color space. Such transformations include converting color images to grayscale for printing, mapping images to the gamuts of target display devices, preparing images for color deficient viewers, and fusing multispectral or multiprimary data into tristimulus images. Each of these transformations has a straightforward, standard mapping, but it often involves a loss of information due to dimensionality reduction or differing constraints for the source and target spaces. In the extreme case, contrast is completely lost when different colors in the source space map to the same color in the target space, an effect known as metamerism. We present a framework for mapping an image from a source color space to a target color space in a way that preserves as much of the local contrast from the source image as possible while staying as faithful as possible to the standard mapping. We adopt a cluster-based approach in which the clusters represent local areas in the image, and the differences between the clusters represent local contrasts between neighboring areas. Our optimization translates clusters optimally to enhance local contrast without making the result seem unnatural. We apply our method to color to gray conversion, gamut mapping, image optimization for color deficient viewers, and conversion of multispectral and multiprimary images to tristimulus images. In each application, our method produces realistic, detail-preserving output images within their target spaces.