Research On Wavelet Transform-Based Scalable Image Coding

The multimedia applications in a multiresolution environment, such as progressive transmission of image, multiquality service and image database browsing, etc., require rate scalable image coding algorithms, and now it becomes a hotspot in image coding domain. In order to improve the decoding quality of images and fit the heterogeneous hierarchy of network, an algorithm of multi-scale transform-based scalable image coding is proposed to solve these problems above.First, this paper summarizes the techniques of the image coding, then gives a summary on the research progress on scalable image coding detailedly, analyzes and compares the advantages and disadvantages of DCT, wavelet and Contourlet transform respectively. Second, the basis of theory for wavelet transform is introduced, which includes the concept of multi-scale analysis, the procedure of Mallat algorithm and the DWT, the local feature of space- frequency, and they are also the necessary base of theory for solutions in this paper. Finally, based on the multi-scale analysis, three embedded quality scalable image coding algorithms are proposed.⑴Based on the zerotree structure, a new predictive coding algorithm for image wavelet coding based on the least square fitting method is proposed which, uses the least square fitting method to predict the wavelet coefficients from the odd subbands to even subbands and achieves better reconstructed quality.⑵On the basis of analyzing the clustering characteristic of significant wavelet coefficients, we use a vector to determine the relative coordinate between two seeds of adjacent significant coefficient clusters. Based on the significant coefficient clusters and the vectors between them, a low bit-rate scalable image coding algorithm is proposed. It uses significant coefficient clusters and vectors between them to effectively reduce the bitrate of encoding the positions of significant coefficients combined with the morphological dilation operator. Experimental results show that the proposed algorithm achieve good subjective quality in low bit-rate, and is superior to state-of-the-art embedded image coding algorithms in terms of PSNR and computational complexity, such as EZW, SPIHT, MRWD, EBCOT, and GW.⑶An algorithm of rate scalable image coding based on vectors between degree-k zerotrees is proposed. It uses a new data structure, degree-k zerotrees, to record the information of relative locations effectively between the important wavelet coefficients of the same direction and the different scales. In low bit-rate, the algorithm can achieve better subjective quality.