一种减少LBG算法计算冗余量的有效改进算法

Efficient Algorithm for Reducing the Computation Redundancy of LBG Algorithm

利用范数等价性和LBG算法迭代过程中数据是分区域收敛性质 ,提出新的码书训练算法FLC。该方法用计算量小的范数进行距离计算和尽早删除聚类过程中已趋稳定的区域 ,因而速度非常快。用典型的测试图像Lena和Barb作实验 ,表明FLC算法以峰值信噪比只比LBG算法少 0 2 5dB～ 0 4 3dB为代价 ,把运行时间缩短为LBG的 1/3 74～ 1/9 5 9。

A fast codebook generation algorithm named fast local clustering algorithm is presented in this paper by the equivalence of norm and the property that data cluster locally. The experiment by the typical test picture Lena and Barb shows that the algorithm is 3.74～9.59 times faster than LBG, and PSNR of it is only 0.25～0.43dB smaller than PSNR of LBG. It is the best algorithm among the six type algorithms introduced in this paper.