【作者】 袁静；

【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， 硕士

【摘要】 对大量图像数据特别是海量的医学影像信息进行高效的存储和传输是当前信息时代所面临的首要问题。除硬件本身的改善外，研究高性能的图像压缩技术变得越来越重要。经过几十年的发展，图像压缩取得了一些成果，并相继制定了一系列标准如JPEG，MPEG和H.26X等，为它的实际应用奠定了一定基础。但这些标准并不完善，特别是对于医学图像的压缩，为了符合临床诊断的某些特定要求，许多新的编码方法仍在研究和讨论之中，分形图像压缩是其中比较有竞争力的一种算法。 本文围绕当今分形图像压缩的热点问题展开研究，分别对灰度图像和序列图像的编码进行了比较细致深入的探讨，研究工作主要包括以下几个方面： 1．针对经典分形编码时间过长的缺点，提出了一种基于小波分解的快速分形图像压缩算法：利用小波分解后各频带间能量分布不均衡的特性，将集中图像主要能量的低频域作为一幅图像进行经典分形编码；利用小波分解后不同分辨率子带图像之间的相似性，用低频域的全搜索代替经典算法中整个区域的全搜索，将低频域得到的分形参数进行适当的比例变换得到整幅图像的分形码。该算法在压缩比不变且基本不影响解码图像质量的前提下大大提高了编码速度，明显改善了分形图像压缩中编解码不对称的现象。 2．对目前基于分类搜索的分形编码算法进行了比较研究，针对这些硬分类方法中类别数需人为预先指定的缺点，采用模糊聚类优化的软分类方法对分形压缩进行了优化尝试。为了保证分类的有效性和准确性，第一军医大学硕士学位论文首先使用LBG(unde一Buz。~Gray)方法对搜索空间组成的样本集进行初始化，然后再根据优化模糊聚类思想(optimai FuZZy CI ustering，oFc)对搜索空间进行软分类，匹配时用类内搜索代替全局搜索，从而减少匹配次数，降低编码时间，使分形编码的效率得到明显提高。 3.对序列图像的分形压缩进行了初步研究，将在静止图像压缩中获得良好性能的模糊聚类优化算法改进并应用于序列图像的分形压缩，充分利用序列图像前后帧之间的相关性有效组织搜索空间，得到了比较满意的编码效果。更多还原

【Abstract】 One primary problem we are confronted with in information age is to effectively store and transmit huge amounts of digital images. This casts increasing the urgency to develop new compression algorithms with high performance. Some popular international standards such as JPEG, MPEG and H.26X are not so perfect and still need to be improved. This is especially true for the grayscale images commonly used in medical areas. Fractal image encoding is one of the competitive schemes in recent research.Focusing on fractal image compression, this dissertation probes intensively into the encoding of grayscale images and sequence images. The main contributions are as follows:1. A fast fractal compression method based on wavelet decomposition is proposed. This algorithm is mainly implemented in two aspects. Firstly, according to the ununiform distribution of the energy in subbands in wavelet transform domain, the low frequency region, where most energy concentrates, is encoded with traditional fractal encoding method. Secondly, with the similarity of subband image among different channels after wavelet decomposition, we have introduced a local optimal search mechanism, instead of global one, to gain the whole image’s fractal codes after suitably rescaling the parameters in low frequency regions. The encoding procedure speeds up greatly while preserving the compression ratio and the quality of the reconstructed image. The dissymmetry between the encoding and decoding process has also been improved dramatically.2. Study also goes deep into some classification based fractal encoding schemes. Taking into account the shortcomings of the hard clustering problem, in which the category number needs to be pre-specified, we aim to modify the fractal image coding algorithm by using Optimal Fuzzy Clustering (OFC) method. Firstly the Linde-Buzo-Gray (LEG) algorithm is realized to get a rough cluster of domain blocks, then optimal category with small computational cost is automatically obtained with the algorithm presented above. Less encoding time is achieved through soft clustering of the domain blocks and classification search. The simulation results show that our algorithm is superior to those fast encoding schemes reported recently.3. Fractal compression of sequence image has also been studied. A fast algorithm is presented in this dissertation. It’s based on the principle of OFC which has been effectively used in still grayscale image compression. The correlation between adjacent images is fully used to organize the domain pool and the coding efficiency is improved greatly.更多还原