A 3D→4D color space transform for efficient lossless image compression

Color images are commonly represented as a combination of different components, for example, Red, Blue and Green in the RGB color space. The components are often correlated and contain common information. By representing an image in a color space in which its components are decorrelated, it can be more efficiently encoded achieving better compression. Most of the approaches to find a suitable color space for image compression are limited to transforming an n component color space to another n component color space. In this paper, we propose a novel transform that converts a 3 component RGB image to a 4 component cGST (color, gray, shade, tinge) image and vice-versa, and show its suitability for image compression. The transform is fully reversible (and hence, is suitable for lossy as well as lossless image compression) and preserves the bit-length for the GST components (allowing existing algorithms to be applied to the components). We develop an encoder-decoder tool using the transform and JPEG-LS prediction scheme, and demonstrate its efficiency (upto 35% better compression ratios over JPEG-LS, 2-5 times less runtime than JPEG 2000 with similar compression ratios) on a diverse set of test images. The transform works especially well for satellite images, computer generated animations and real images with shadows. The work also opens the scope for studying color transforms not restricted to matrix multiplication or n→n dimensional conversions for image compression. Our work also adds to the understanding of the impact of shadows on color components and is useful in image analysis in general.