摘要
本文提出了一种高质量的VQ初始码书生成方法一自适应决策导向聚类算法。该算法通过自适应地调整保护区的上下限,逐渐确定聚类的最小失真阈值和形心间的最佳距离。实验结果表明:生成的初始码书的典型性较好,在后期LBG训练中收敛较快,在码率为0.5bpp情况下,重构图像的峰值信噪比为30～36dB,建议算法的码书性能优于其它方法。
A new method of high-quality VQ initial codebook generation-Adaptive Decision-Directed Clustering (ADDC) is presented. Through adaptively adjust the upper and lower limit of guard zone, minimum distortion threshold among clusters and optimal distance between shape-centers are determined. The experimental results show that the generated codebook has good typicality and fast convergence speed in the following LBG train, and the proposed algorithm has a better performance than other methods, and a PSNR of reconstructed image is 30-36dB with a rate of 0.5bpp.
出处
《模式识别与人工智能》
EI
CSCD
北大核心
1998年第4期378-384,共7页
Pattern Recognition and Artificial Intelligence
基金
国家自然科学基金
电科院预研基金
关键词
图像编码
矢量量化
初始码书
LBG算法
Image Coding, Vector Quantization, ADDC, Initial Codebooli, Guard Zone
