Research On Content-based Embedded Image Wavelet Coding Algorithms

With the development of multimedia technology, image information gradually becomes one of the main measures of communication. How to resolve the problems of images transmission and storage is the focus in image compression. From JPEG, MPEG-1 to JPEG-2000, MPEG-4, technologies of image compression have been developing rapidly. At present, embedded image coding algorithms based on wavelet analysis are dominant, such as EZW, SPIHT, EBCOT and LZC. Research on content-based embedded image wavelet coding algorithms concentrates on the following aspects: how to describe image content correctly and extract shape feature efficiently; how to organize coefficients to improve the efficiency of coding; how to quantitate coefficients to utilize context relativity and make quality of images accorded with HVS etc.In this thesis, firstly, the theory of image compressing is introduced. The methods of embedded coding and international standard of image, video coding are also be listed. The thesis researched the measures of image content description based on the feature of color, texture, shape and space. Considering the key feature of image content, an invariant moment based on threshold optimization method is used to extract the shape feature. The experiment results demonstrate that the method hold the shape consistency of image well, it also describe the shape information of image completely.Secondly, the thesis introduces the theory of image wavelet transform, the concept of multi-resolution analysis and the Mallat arithmetic by which the signal is fast composed and decomposed. Combined with the wavelet lifting scheme which fit embedded image compression, the characteristics of the wavelet transformed image coefficient distribution, gray level histogram and the image itself are further analyzed, and the reasonable wavelet bases for image compression are obtained.In addition, the principle and coding course of EBCOT algorithmic have been studied, which is the core coding algorithmic of JPEG2000 standard. In order to resolve the algorithmic complexity and reduce the time of coding and decoding, a measure which saving row-status in advance to separate the relationship of strip and encoding logic contexts based on single window and computing bit-plane status in priority, a parallel EBCOT encoding algorithm is proposed to encode bit-plane and its three passes in one time scanning.Finally, the typical embedded coding algorithms and schemes are analyzed, such as EZW, SPIHT, EBCOT, 3D-EZW, 3D-SPIHT and 3D-ESCOT. A new context model of wavelet coefficients is proposed, which make use of partial order, self-similarity between subbands of wavelet coefficients and the correlation of neighboring wavelet coefficients. Experimental results show that the model achieves higher PSNR than 3D-ESCOT while keeping all the advantages of 3D-ESCOT.