Image compression using non-seperable statistically matched wavelets

Traditional image coding schemes employ discrete cosine transform based coding techniques for image compression. The very recent JPEG 2000 standard uses discrete wavelet transform with a separable forward transformation kernel. This approach uses a standard CDF 9/7 wavelet for compressing images. In this paper, we address the problem of non-separable matched wavelets for image compression. The image is modeled as self similar process in which the analysis filters (scaling and wavelet filter) are matched to the statistical properties of the image. The filters which are matched to the image at the first level of decomposition are then used for multilevel decomposition. Results have been evaluated using these approaches in the form of auto-correlation plots and Histogram plots. These experimentations have yielded encouraging results.