A Wavelet Image Coding, Based On Predicted Classification

Abstract Today multimedia technology is the hot issue in the scientific research, and the core of that is image coding. Because more definit image need more disk space to store and more bandwidth to transfer, image coding is absolutely necessary to save this cost. However, image coding must try to avoid image distortion, in other words, the reconstructed image should almost as same as the original one. This need lossless or high fidelity image coding technology. To the former, it can only compress image 3 or 4 times and this can't meet the request of application. Therefore, studying the fidelity image coding is significant. Wavelet transform is the new tool applied in image coding field. How to organize the wavelet coefficients effectively is pivotal to the performance of image coding based on wavelet transform. One typical example of this aspect is the Zerotree. It takes the advantage of parentæ¢'hild dependencies of zero coefficient between the different wavelet subbands, and represents nozero coefficients indirectly by focusing on zero regions. Zerotree arbitrarily shaped zero regions as the union of a highly constrained set of tree structured regions. As a result, certain zero regions not well aligned with the tree structure may be expensive to represent, while other portions of zero regions may not be included in Zerotrees at all. Furthermore, because Zerotree represents significant coefficients indirectly, it will decrease the quality of reconstructed image. In this paper, a predicated classification model for image coding is presented by studying of wavelet coefficients statistical properties. Contrary to the Zerotree based method, this model takes the advantages of dependencies between significant coefficients in the wavelet domain. By using this model combined with the proposed seed-dilation algorithm, our experiment show that the results of image coding have better reconstruction quality and high robustness.