Image Compression Based On Adaptive Transform

Digital images have been playing an increasingly important role in information areas. Image compression technique can help to reduce the burden of storage and transmission of the huge image data volumes. Moreover, with the development of computer and electronic technology, the digital image has infiltrated all areas of life and production. Image compression is the first process step during image acquisition, storage, transmission and display, and the subsequent processing and use largely depend on the quality of image compression.The theory and techniques for image compression are with the theme of disciplines of adaptive reversible integer transform. Starting from the principle of image segmentation, the paper first makes detailed the basic principles and methods of quadtree block. Then, we make detailed analysis about some transform methods which are always used in image compression based on the basic discipline of floating-point transforms. And we researched about the principles of reversible integer transform, which forms the basis of the following study. Finally, based on time-domain lapped transform and quad-tree sub-blocks, we realize adaptive compression for two-dimensional images under the lossy and lossless compression conditions, respectively.In lossy impression respect, we propose a progressive adaptive compression method for two dimensional images based on hieratical time domain lapped transform and quadtree block. Compared with conventional fixed blocked DCT-based algorithms, the improvements consist of two aspects. Firstly, we reorganize blocky transform coefficients into two-dimensional wavelet subband structure, making transform coefficients of the same frequency concentrated in the same subband. Secondly, we encode transformed coefficients using set partition embedded block algorithm which proves to be an efficient codec with low complexity.In terms of lossy-to-lossless compression, we study about two-dimensional adaptive compression methods based on reversible integer lapped transform. Our work includes two aspects.(1) A new lapped transform(RLT) is proposed based on time domain lapped transform model. Pre-filters are used in RLT to exploit correlations between neighboring blocks, shifting high-frequency components to low-frequency; and these components are reconstructed by post-filters in the inverse transform. In this way, blocky transform efficiency is improved and blocking artifacts are reduced. Basic matrix of lapped transform is decomposed using matrix factorization method, and reversible integer transform is realized by multi-lifting structure. Combined with hierarchical structure, we propose a new compression algorithm for two-dimensional images with comparable even better performance than JPEG2000 in lossless compression.(2) Elegantly integrated with quadtree block algorithm, the algorithm is able to get several blocks with different size, according to appropriate threshold block chosen based on image texture analysis, which takes full account of the characteristics of image to remove the correlation between pixels.