Research On Low Bit-Rate Image Coding And Denoising Algorithm

Image compression encoding and image denoising have always been important and hot research problems all along in signal procession technologies. Wavelets have found wide applications in image proeessing due to the time-frequenecy localization, enegry compaction and multiscale decomposition property. It has been adopted as a key technique in the new still image comperssion stnadard JPEG-2000 and image denoising area. Embeded Zerotree Wavelets (EZW) coding algorithm not only has high compression rate but also has some new features such as:progressive coder, and low computational complexity. SPIHT(Set Partitioning in Hierarchcal Trees) is one of the excellent typically algorithm. However, in high dimensions (such as two dimensions), wavelets expose an inherent limitation. It is because wavelets are optimal bases in catching zero-dimension singularities but not for one-dimension or higher dimensions singularities. Recently, researches about Multiscale Geometrie Analysis (MGA), which have the more powerful directional and the more sparseful approximating multivariate function in high-dimension space than wavelet analysis, gradually are accepted. Wavelet-based embedded zero-tree coding improvement and multiscale analysis and its application in image ROI coding and denoising are investigated in detail in this dissertation.The main work can be summarized as follows:Firstly, it offers an embedded coding based on zero-tree cluster under the condition of lower bit-rate and complexity—embedded zero-tree cluster encoding (EZSC). The algorithm is an improvement of SPIHT. It analyzes the distribution characteristics of wavelet coefficients as well as quantitative planar features through a large number of experiments, and do a deep research to the effective organization of wavelet coefficients, adaptive quantization bit-plane setting, coefficient scanning order and so on. The algorithm proposes zero-tree cluster conception, uses information theory to analyze the reliability of embedded zero-tree cluster encoding, and puts forward the improvement direction of the algorithm. Finally, due to zero-tree cluster conception, the algorithm combines breadth-first and depth-first scanning code, together with using step quantitative strategy to design EZSC coding, which enhances the efficiency of embedded coding under of condition of low rate.Secondly, the algorithm introduces wedgelet transformation to the handling of image background region, and enhances embedded coding functionality of interested regions. It makes use of the ability of wedgelet transformation to image border's better sparse approximation, and firstly make Wedgelet transformation preprocessing to the background region of images to lower background information on the region entropy. Secondly recombines background regions and interested ones, uses EZSC to do Synthetic images an embedded coding after wavelet transform and bit plane upgrade. The result shows that the algorithm increases PSNR values of interested regions in low rate conditions, and the qulity of the whole reconfigurable images decrease within limits, thus make a coordinated distribution between the quality of interested regions together with background ones and rate in low rate conditions.Finally, the algorithm proposes multiple transform domain image denoising idea based on wavelet transform domain and Contourlet transform domain. It uses the advantages of wavelet's optimal indication to point singular signal in images, combines with the relationship of the similarity of wavelet tree and Hidden Markov Tree, denoises the images at the first level. Secondly, it uses the feature that contourlet transformation can get a better expression ability to high-dimensional image signal than wavelet transformation, so the algorithm adopts contourlet transformation to sparsely and approximately denotes the first processed high-bit image signal, further removes the image noises through setting different threshold processing of multiple iterations. The method could improve the visibility and fidelity of reconfigurable images. The result shows that the proposed method gets an improvement in respect of, such as PSNR values, the visibility of reconfigurable images and so on.