Research On Contourlet-Based Scalable Image Coding Techniques

In many applications, due to varied user requirements, terminal abilities, quality of services (QoS) provided by heterogeneous networks and changes of network transmission, such as noises and packet blocking, different image quality is demanded. The best solution to this problem is scalable image coding.This paper first summarizes the research progress on scalable image coding based on DCT, wavelet and Contourlet transform respectively, then analyzes and compares the advantages and shortcomings of three kinds of algorithms above. Second, the principle and the procedure of Contourlet transform are introduced. Through experimental statistics and theory analysis, the distribution characteristics of Contourlet coefficients is detailed and modeled by inter-scale spatial orientation tree structure and intra-subband energy block and zero-block. Finally, based on the distribution model of coefficients, this paper proposes three embedded quality scalable image coding algorithms.⑴Based on the spatial orientation tree structure, a Contourlet-based SPIHT (CSPIHT) algorithm is proposed which, besides the features of wavelet-based zerotree coding algorithms, has the characteristics of multi-directionality and anisotropy. Especially, the proposed algorithm can efficiently capture the edge and texture information of images so as to obtain good visual quality. Experimental results show that the PSNR of reconstructed images by CSPIHT is higher than the one by SPIHT at low bit rates. And at medium bit rates the PSNR is slightly lower than the one by SPIHT, while the visual quality of reconstructed images by CSPIHT is better.⑵Based on the distribution characteristics of significant Contourlet coefficients, this paper proposes a multi-scale successive approximation quantization scheme. Combined with CSPIHT, the algorithm is superior to the traditional SPIHT both in terms of PSNR and the subjective quality of textures and contours in the decoded image.⑶Using the energy block and zero-block model, a novel image coding algorithm based on quad-block partitioned strategy is proposed. In this algorithm, the side information of each coefficient is replaced by the information of its block to reduce the cost of coding side information. And the quad-block partitioned strategy further makes use of the relationship between neighboring coefficients inside the same direction subband to efficiently organize significant coefficients. Experimental results prove that compared with SPIHT, this algorithm not only can obtain better subjective quality, but also get 0.5~1.0dB higher PSNR for images with temperate textures, and even 1.0~1.5dB higher PSNR for images with rich textures and edges.