Adaptive Color Transformation Algorithm Based on Color Channel Correlation

To achieve efficient color image compression, it is essential to fully exploit the inherent correlations among color components, which is typically accomplished by transforming RGB components into a decorrelated color space representation. An adaptive color transformation algorithm (ACT) based on color channel correlations is proposed in this paper. The inter-channel redundancy in RGB images is reduced through linear fitting methodology according to the correlations between color channels. The relationship among linear fitting coefficients, variance of fitting residuals, and inter-channel correlation coefficients is theoretically derived, with the transformation matrix being optimized following the minimum variance principle. To address the issue of varying fitting coefficients across different color regions within an image, an adaptive block division algorithm is further proposed to enhance the ACT framework. Comprehensive evaluations are conducted on both the Kodak test suite and a custom high-resolution natural scenery image dataset. Experimental results demonstrate that the proposed ACT algorithm outperforms conventional color space transformation models including RCT and YCgCo, as well as existing improved algorithms.