On Rate-Distortion Models for Natural Images and Wavelet Coding Performance

摘要: Operational rate-distortion (RD) functions of most natural images, when compressed with state-of-the-art wavelet coders, exhibit a power-law behavior DpropR-gamma at moderately high rates, gamma being constant depending on the input image, deviating from well-known exponential form RD function Dprop2-xiR for bandlimited stationary processes. This paper explains this intriguing observation by investigating theoretical and operational images. We take as our source model fractional Brownian motion (fBm), which is often used to nonstationary behaviors in first establish that fBm process (both 1-D 2-D) indeed follows power law. Then we derive encoded based water-filling principle. Interestingly, both behave DpropR-gamma. For values are found be distributed around 1. These results lend an information support merit multiresolution compression self-similar processes and, particular, images can modelled such They may also prove useful predicting performance optimized image coders