A Comparison of Entropy ConstrainedStill Image Subband Coding and Transform

The paper compares frequency domain coders employing two entropy coding schemes: (1) Adaptive classiication of the frequency components followed by entropy coding of each class of components and (2) straightforward, nonclassiied entropy coding of the frequency components. Both schemes use uniform threshold quantization, and the entropy coders are nonadaptive, multialphabet arithmetic coders. To optimize the arithmetic coders in the rst scheme, an approach based on modeling the frequency components as a mixture distribution is used. The eeciency of this model is demonstrated through theoretical and practical coding results. The schemes are compared using both a DCT and a subband lter bank for signal decomposition. It is concluded that the adaptive classiication coder model gives a signiicant rate gain as compared to a nonadaptive coding scheme, and that subband ltering outperforms the DCT.