Adaptive Image Coding Based on Matching Pursuit

A new image coding method is proposed that can produce an embedded and progressive-PSNR bit stream with region of interest ability. It represents the original image as linear composition of some time-frequency atoms. These atoms are extracted from a redundant dictionary to match the time-frequency characteristics of original image in a great degree. Our experiments show that it can capture texture and edge characteristics efficiently and the quality of recovered image is better than SPIHT at low bit rates.