A Quality Scalable Coding Algorithm of Images at Low-Medium Rate Based on Contourlet Transform

In this paper, a new non-linear image approximation method that decomposes images both radially and angularly is proposed. In order to explore the potentiality of the Contourlet transform as a tool for image coding, SPIHT coding scheme is developed into a non-linear image approximation technique. Through careful statistical analysis of the independent sub-band coefficients of Contourlet transform, a novel directional tree structure is proposed, and the “zero-tree” characteristic of this structure is statistically validated. According to the distribution characteristic of the “significant coefficients” of Contourlet transform in different sub-bands, a multi-scale successive approximation quantization scheme is proposed, which takes the anisotropy characteristics of each Contourlet sub-band into consideration so as to effectively enhance the capability of capturing textures, contours and fine details in images. Based on the directional tree structure and the quantization scheme, a novel embedded quality scalable image coding algorithm based on Contourlet transform is proposed. This algorithm not only has the useful characteristics of all the wavelet-based zero-tree coding algorithms, but also can more efficiently capture the direction and anisotropy features usually represented in natural images than wavelet transform. Experimental results show that at the low to medium bitrate, the decoded image of the proposed algorithm is superior to that of the traditional SPIHT coding algorithm both in terms of PSNR and the subjective quality of textures and contours in the decoded image.