Research On Image Coding Based On Directional Multi-scale Transform

Image coding as the key technique of modern communication, media storage, data distribution and multimedia computers is one of the most active areas in information processing.The last decade, wavelet transform has become a powerful tool to analysis and process natural images.However, the capacity in the sparse representation of wavelet is relatively limited in two-dimensional image, or higher dimensional data. Because the isotropic wavelet basis function can not be effective use of the geometric characteristics of the data itself to capture the image in the direction of the line along the edge singularity.To some extent, it limits the efficiency of wavelet-based image coding.To overcome the lack of wavelet transform, at present, based on the theory of wavelet transform, there are a series of new and more effectively mathematical transformation to express and deal with the singular of high dimensional data. They not only have multi-resolution features,time-frequency localization, multi-directional and anisotropy,but also overcome the limitations of wavelet transform in the representation of the edge, high profile and high-dimensional singular. Wavelet transform based on adaptive characteristics has been got more attention.First of all,this paper systematically introduces the principle of image compression and the research of directional multi-scale geometric analysis,makes a comparison of the directional multi-scale geometric analysis tool.Second, we construct a directional multi-scale geometric analysis tool named directional adaptive lifting wavelet (DAL-DWT), the program can overcome the shortcomings that conventional DWT is difficult to approach the edge of the image in any direction and texture; according to the image minimum prediction residual energy of the block,which can adaptive adjust the direction of the filter, better captures the texture feature information, more effectively eliminate the redundancy between neighboring pixels;reduce the high frequency energy of coefficients,low-frequency has better energy concentration correspondingly; without the directional block segmentation of rate distortion optimization, adopt the fixed directional block size and prediction mechanism of directional information, reduces the overhead of side information, these aspects lay the foundation for improving the image coding efficiency.Then, after experimental statistics and theoretical analysis to the each DAL-DWT sub-band coefficients, we proposed two kinds of quality scalable embedded image coding algorithm successively:(1) embedding zero-tree coding algorithm in the directional adaptive lifting wavelet domain:analysising statistically the distribution characteristics of DAL-DWT coefficients, which include very good spatial and frequency domain localization properties, the good "zero tree" feature, and compared with the traditional lifting wavelets, further applied the DAL-DWT to zero-tree coding structure of image can obtain better coding efficiency and visual effects than traditional lifting wavelet transform, the effect of image with rich texture information becomes more apparent especially; (2) a kind of region of interest (ROI) image coding scheme:we propose that selecting important sub-band and shifting by ROI mask and the energy based on the DAL-DWT, simultaneously expanding the important sub-band which should meet the "zero-tree" characteristic, a new region of interest(ROI) coding method based on selecting sub-band and bitplane shift is proposed. Experimental results validate the validity of proposed algorithm.Finally, the paper proposes a image coding method based on Directionlet transform. Directionlets not only has the characteristics of separable wavelet transform filtering, twice sampling and simple calculation, but also has more flexible ability to capture the anisotropy than the contourlet transform.The paper proposes the spatial orientation tree structure in transform domain. A large number of experiments prove that the tree structure meet the zerotree properties.On this basis, the paper proposes a SPIHT coding method based on Directionlets suitable for medium and low bit rate.Experiments show that the quality of the reconstruction image of proposed method is better than SPIHT method.Especially in low bit rate conditions,the decoded image captures more directional information of image and preserves the edge information in all directions.