Supra-threshold perceptual image coding

Abstract

We investigate different algorithms and performance criteria for supra-threshold image compression. The algorithms include JPEG and perceptual JPEG, the Safranek-Johnston perceptual subband image coder (PIC), and the Said-Pearlman algorithm which is based on Shapiro's embedded zerotree wavelet algorithm. We also consider a number of performance criteria. These include mean-squared error, Watson's perceptual metric, a metric based on the PIC coder, as well as an eye-filter weighted mean-squared-error metric. Our experiments indicate that the PIC metric provides the best correlation with subjective evaluations. The metric predicts that at very low bit rates the Said-Pearlman algorithm and the 8/spl times/8 subband PIC coder perform the best, while at high bit rates the 4/spl times/4 subband PIC coder dominates.