Spatial analysis-synthesis for improvement of wavelet coders

This paper describes an adaptive image compression scheme built upon a conventional wavelet coder. The principle is to warp the input signal so as to minimize its coding cost. First, the warping parameters are estimated through an analysis step. Then, the warped image is sent to the basic coder. The warping parameters are transmitted independently. Finally, a synthesis step reconstructs the original image by inverse warping. This approach of the adaptivity issue is different from previous methods seeking to adapt the wavelet rather than the signal. Our technique enables to take advantage of optimized wavelet coders such as JPEG2000. Moreover, the minimization procedure does not make any assumption about the regions of high coding cost. An implementation of the principles is presented where the warping transform is modelized by a regular 2D-Mesh. Compression results of the Analysis-Synthesis (A-S) scheme show a significant visual quality improvement for non-texture images compared to JPEG2000.