On the perceptual interband correlation for octave subband coding

Abstract

The ability of human vision systems to distinguishing gray values of pixels depends on factors such as background luminance, the nearby image contents, etc. Thus, for any image, there are various representations subjectively judged to be the same with the original. This subjective phenomenon provides the conjecture that some of the various representations may provide the statistical interband correlation instead of just the perceptual similarity with the original image. This conjecture motivates the work to exploit the desired representations and check the associated new merits. This paper demonstrates that the most desired representation can be systematically exploited such that the interband correlation of octave subbands is increased. We present two methods with different complexity and show the resultant coding gains. The gains indicate that there exist amounts of perceptual correlation applicable to image coding despite that the original interband correlation has been considered low.