基于最小描述长度的Context量化方法研究  

Research on Method of Context Quantization Based on Minimum Description Length

作　　者：卜春芬[1] 陈旻[2] 殷启新[2] 

机构地区：[1]昆明学院物理科学与技术系,云南昆明650214 [2]云南警官学院信息网络安全学院,云南昆明650223

出　　处：《昆明学院学报》2017年第6期63-67,共5页Journal of Kunming University

基　　金：国家自然科学基金资助项目(61062005);云南省自然科学基金资助项目(2016FB113)

摘　　要：介绍一种自适应优化Context量化算法.量化器优化目标为令训练序列描述长度最短.定义描述长度增量作为聚类相似性测度并使用混合聚类算法实现Context量化以保证量化器在多进制信源下获得优化量化结果,从而克服以往优化Context量化器不能应用于多进制信源或不能自适应获得量化级数等限制.最后将量化算法应用于图像小波压缩.实验结果表明,最短描述长度Context量化器能够获得与人工精心调试的经验量化器类似的压缩效率,而不依赖于人工经验.The self adaptive optimized context quantization algorithm is presented and the optimized aim of the quantizer is to minimize the description length of the training sequence. The increment of the description length is defined as the simi larity measure and with the clustering operation, Context quantization is realized to guarantee that the quantizer gets the optimized and quantitive results under the multi-system information source so as to overcome the limits of previously proposed context quantization algorithms which cannot be ap-pl ied to non-binary sources or cannot be adaptive to determine the quantization levels. The proposed algorithm is then applied to an im-age wavelet compression system. Experiment results indicate that without any human intervention, context quantizer of minimum de-scription length can acquire the simi lar compression efficiency as the empirical quantizer careful ly tweaked by people.

关 键 词：Context量化 描述长度 描述长度增量 图像小波压缩 

分 类 号：TN919[电子电信—通信与信息系统]