【作者】 胡宏梅；

【导师】 董恩清；

【作者基本信息】 苏州大学 ， 通信与信息系统， 2007， 硕士

【摘要】 矢量量化是数据压缩的重要关键技术之一,它主要包括三方面研究内容:码书设计、码字搜索和码字索引分配。而码书设计又是矢量量化研究中需要解决的最关键的问题。文中主要分别对码书设计中的LBG算法中空胞腔处理方法研究、蚁群码书设计算法研究、粒子群码书设计算法研究及基于局部余弦变换的分维矢量量化四个问题展开研究。首先,在分析了处理LBG算法中空胞腔的次邻域方法的基础上,针对次邻域法所存在的缺点,本文提出了一个新的空胞腔处理方法,结合最大胞腔分裂法,采用仅聚类产生最小失真的次邻域矢量的方式来处理空胞腔,减少量化失真,改善量化性能。其次,在基于蚁群算法的码书设计研究中,本文提出了将频率敏感算法引入到基本的蚁群算法中,通过增加失真测度来减小蚂蚁重复选择同一胞腔的可能性,从而有效地提高了其全局搜索能力,避免了“早熟”现象。再次,在基于粒子群算法码书设计研究中,提出采用随机概率扰动的方式作为基本粒子群算法的全局极值更新条件,从而增加全局最优区域的搜索,避免了粒子过早的“趋同性”。最后,在基于局部余弦变换的分维矢量量化的研究中,对LCT变换系数采用分维矢量量化的方式进行码书设计。分别从客观评价和主观评价两方面来验证提出的分维矢量量化算法的有效性。更多还原

【Abstract】 Vector Quantization (VQ) is a significant technique in image compression. VQ research includes codebook design, codebook search and codebook index assignment. And codebook design is a key technique of VQ.In this paper, empty voronoi problems, ant colony codebook design algorithm, particle swarm optimization codebook design algorithm and split vector quantization based on Local Cosine Transform (LCT) will be discussed.First, Hypo-adjacent technique is a way to solve the empty voronoi problems. To overcome the limitation of the technique, an improved technique that is based on the technique is proposed. The improved technique combines the largest voronoi splitting technique, and adopts the way to only cluster the Hypo-adjacent vector with the smallest distortion measure to deal with the empty voronoi. As a result, total distortion measure is decreased, and the performance of vector quantization is improved.Second, in ant colony codebook design algorithm research, frequency sensitivity measure is adopted in the original basic ant colony algorithm. Since the scheme decreases the probability of choosing the same cell by increasing the distortion measure, the randomization of choosing cell is increased. The scheme can avoid receiving a local optimization solution, and enhance the global search ability.Third, in particle swarm optimization codebook design algorithm research, the global best position updating condition in basic particle swarm optimization algorithm is modified. Random probability condition is adopted to update the global best position, extending the global search area and avoiding the premature phenomenon.Finally, in split vector quantization based on LCT research, split vector quantization is adopted to quantize LCT transform coefficients and to design the codebook. The availability of improved split vector quantization algorithms is validated from the objective index and the subjective index, respectively.更多还原