Adaptive Technology Is Applied Research In Image Compression

Adaptive techniques always play an important role not only in still image compression but in motion image compression standard. Although much studies in adaptive coding have been done in recent tens years, many of these studies always lead to discontinuity of the adjacent blocks and block effect, because traditional DCT coding depends too much on statistical model in classified adaptive quantization for image blocks, thus give rise to improper quantization of some blocks. In this thesis, the structure characteristics of the image are analyzed and utilized to segment the image adaptively.Firstly, an adaptive segmentation DCT coding method based on edge information measure is developed. An image is recursively split into different size leaf-blocks, the edge information measure based on the structure characteristics is adopted as the splitting criterion. Each leaf-block will be quantized by different quantization template in transform domain.Then, to eliminate the additive bit-consumptions recording the types of blocks, the block-merging transmission mode and the adaptive decoding based on the DCT spectrum feature of the blocks are developed. The simulation experiments on computer prove that this method has higher coding efficiency than traditional DCT methods.Secondly, an adaptive block segmentation DCT coding method based on the DCT spectrum main orientation is developed. The former study showed that the directional edge has orientation in its DCT spectrum distribution and developed the concept of the DCT spectrum main orientation. Based on this concept, the smooth and the edge field can be segment adaptively, then coded by DCT separately. This method takes full advantage of the concept of spectrum main orientation which can not be utilized very well in the non-adaptive segmentation method, and it is proved by the simulation experiment that this method enhances the compression ratio, and saves the runtime.At last, the adaptive vector quantization coding based on the DCT spectrum main orientation mentioned above is developed. Utilizing DCT spectrum main orientation, the input image can be classified. And then different codebooks are designed responding to the different kinds of blocks. Each codebook has smaller size than one codebook in unclassified vector quantization, so the runtime can be decreased. It is proved that this method has a higher PSNR ,and if a optimizing condition, which is developed in this chapter, is satisfied, the problem of lower compression ratio caused by multi-codebooks can be solved. At the end of this chapter, the performance and availability of the VQ based on some other adaptive block segmentation methods are discussed.