基于SOFM和LVQ神经网络的矢量量化图像压缩.docVIP

基于SOFM和LVQ神经网络的矢量量化图像压缩 摘 要 矢量量化作为一种高效的数据压缩技术，已被广泛应用于图像压缩领域。本文在系统介绍矢量量化数据压缩理论的基础上，着重探讨了矢量量化的关键技术——码书设计算法： 回顾了经典的矢量量化码书设计算法——LBG算法，研究了基于竞争型神经网络SOFM和LVQ的矢量量化码书设计算法； 通过仿真实验，采取将图像分块构建矢量的思想，研究了在改变码字大小等参数的情况下，这三种算法的训练过程和算法性能； 根据LBG算法的特性，提出一种联合SOFM和LBG算法的码书设计改进方法，并通过实验验证了改进算法的有效性。 大量的实验结果表明，LBG和LVQ算法对初始码书依赖性大，SOFM算法训练时间较长。对于给定的码字大小，码书越大，压缩比越低，但重建图像质量越好；码书相同时码字较小的编码性能较优。 关键词：码书设计；LBG算法；SOFM神经网络；LVQ神经网络；改进算法 The Vector Quantization in Image Compression Based on SOFM and LVQ Abstract As an efficient data compression technique, vector quantization(VQ) has been widely applied to image compression. This paper presents a systematic introduction to the theory of VQ data compression, focusing on the key technology of VQ——codebook design algorithm: The classical algorithm of VQ codebook design—LBG algorithm is reviewed. The VQ codebook design algorithm based on the competed artificial neural network—SOFM neural network and LVQ neural network are studied. The training process and performance of these three codebook design algorithms are studied through simulation experiments. During the experiments, image blocking is adopted in the construction of the vectors. An improved method for codebook design based on the combination of SOFM and LBG is put forward according to the characteristics of LBG algorithm and its validity is validated by the experiment. The experimental results indicate that the LBG algorithm and LVQ algorithm are sensitive dependence to initial codebook; SOFM algorithm takes a long training time. For a given size of codeword, a larger codebook will result in a lower compression ratio but a better quality of reconstructed image. When using the same codebook, the coding efficiency of less codeword performs better. Key words: Codebook design; LBG algorithm; SOFM neural network; LVQ neural network; Improved method 目 录 第1章 前 言 1 1.1 图像压缩的目的和意义 1 1.2 各种图像压缩方法介绍 2 1.3 矢量量化技术的研究现状 3 1.4 本文的研究目标和主要内容 4 第2章 矢量量化基础知识 6 2.1 矢量量化的基本原理 6 2.1.1 矢量量化的理论基础 6 2.1.2 矢