【作者】 冯前进；

【导师】 陈武凡；

【作者基本信息】 中国人民解放军第一军医大学 ， 生物医学工程， 2003， 博士

【摘要】 近十年来，随着小波理论研究的不断深入，基于小波变换的图像压缩的成为图像编码的研究热点。在此背景下，本论文进行了基于小波变换的医学图像编码的研究，旨在提出新的图像编码方法。论文的主要研究内容如下： 1)提出了基于小波变换的块优化等级树图像编码算法，该算法将图像分块处理，块内编码采用一种新等级树算法，较传统的SPITH方法增加了预测过程，进一步利用了相邻小波系数的相关性。块间根据率失真曲线对比特分配进行合理优化。同时，由于数据是分块编码的，本算法较SPITH有更好的容错能力，且易实现感兴趣区(ROI)编码。 2)提出了基于小波变换的预测四叉树图像编码算法，该算法属于带内编码与带间编码的混合。将同一频带内系数分块，随比特面的移动，将块由大到小进行四叉树分裂，克服了固定大小块的不足。另外在编码过程中加入了预测过程，实现对块的裁剪，以使块的形状更符合实际的情况。熵编码采用了基于上下文的算术编码，提出了四种上下文编码模型。通过对比实验表明，该方法的压缩性能较EZW、SPIHT、JPEG2000、QT_L均有不同程度的提高。 3)提出了两种新的模糊矢量量化算法，第一种算法是通过直接改造相容关系矩阵得到等价关系矩阵，从而克服闭包运算对相似准则的放大作用。第二种算法根据率失真曲线对FC树进行优化裁剪，较传统的FC更为合理。对比实验显示，用新算法得到的解码图像在信噪比和视觉评价上较LGB算法都有显著提高。第一军医大学博士学位论文 4)提出一种基于小波变换的内嵌矢量量化图像编码方案。其编码模型借鉴了SPD注T算法的思想，同时结合多级矢量量化，实现了内嵌编码并得到高的编码效能。尤其对于医学类型图像的压缩，有优异的表现。更多还原

【Abstract】 More and more researches focus on wavelet based image coding as the wavelet theory research proceeds in last decade. Under this background, my research is mainly about the development of novel wavelet based image coding schemes with special reference to medical images. In this thesis, several new medical image coding algorithms are put forward. The main contributions are as follows:1) Proposing a new EBOHT(Embedded Block Coding with Optimized Truncation of Hierarchical Trees) image coding algorithm. In this scheme, Wavelet coefficients are divided into many blocks according to its spatial position. For each block, the algorithm adopts a improved SPIHT which has a predictive procedure to exploit the correlation among the adjacent coefficients. Then a optimizing method based on rate-distortion curve is used to make reasonable bit-allocation among the blocks. By arranging the datum into blocks, EBOHT outperforms SPIHT with regard to error resilience and is easy to achieves Regions of Interest (ROI) coding of images.2) Proposing a new Wavelet Image Coding method based on Predictive Quadtree-split. This method hybridizes the intra-band and inter-band coding. Wavelet coefficients in same band are partitioned into many blocks, and the blocks split according to quadtree with the moving of bitplane. This kind of dynamic block spliting can overcome some shortcomings brought by fixed block size. At the same time, the algorithm introduces a predictive coding procedure to truncate the block and make the block’s shape fit in with the actual image. A context based arithmetic coder is adopted in entropy coding, and four different sets of models are designed for the arithmetic coder. The experiment results show that the proposed algorithm provides better compressionratio compared with EZW, SPITH, JPEG2000, QT_L.3) Proposing two kinds of new fuzzy clustering algorithm. The first one gets the equivalent matrices by modifying the fuzzy consistent matrices directly. So it can avoid the expending of distance measures. The second one truncates the FC-Tree based on rate-distortion curve to get a optimal clustering. The objective and subjective evaluation shows that these two algorithms provide significant improvement compared to LBG.4) Proposing a new VQ compression scheme. The method realizes the embedded coding through MVQ and SPIHT like coding structure, the experiment results shows this scheme can get excellent performances when using the corresponding codebook for specified type of medical images.更多还原