Correlation VS. decorrelation of color components in image compression — Which is preferred?

Most image compression techniques are based on de-correlating the 3 color primaries before applying monochrome encoding to each of the resultant 3 components. Recently, however, a new approach to color image compression based on exploiting the correlations between the color primaries has been introduced. This method, in several cases, outperforms the common de-correlation approach. In this work we analyze and compare the two methods in their optimized cases using a Rate-Distortion framework for subband transform coders. Our goal is to provide theoretical tools to predict which method yields the best performance for color image coding. Our results show that both methods outperform JPEG. However, theoretically and perceptually, the correlation based technique should be preferred. Our conclusions are instrumental in the decision of which color compression method to select in various applications of limited bandwidth communication.