Reversible Color Spaces without Increased Bit Depth and Their Adaptive Selection

The efficient compression of color images requires a processing step exploiting the correlation between the color components. This is typically realized using a color transformation. In lossless compression systems, the reversible color transformation increases the bit-depth for chrominance components from eight to nine bits per pixels. This can be avoided by using modulo arithmetic, while keeping the property of reversibility. This letter investigates the impact of these modulo operations on the compression performance, compares different processing structures, and proposes a new adaptive selection of suitable color spaces. It is shown that (i) the limitation of the bit depth generally leads to lower compression performance, (ii) the drop in performance depends on the processing structure used, and (iii) the average performance can be improved by the proposed adaptive selection of the color space.