一种不需要计算量化失真的快速码书训练算法

A FAST CODEBOOK TRAINING ALGORITHM WITHOUT THE COMPUTATION OF DISTORTION

该文利用LBG算法迭代过程中质心序列收敛特性,提出了一种快速算法。它的基本思想是,直接去掉LBG算法中量化失真计算,用质心序列收敛作停止条件。我们用典型的测试图像Lena做实验,实验结果表明,该算法与著名的LBG算法的PSNR相差小于0.1dB,但它的运行时间至少比LBG的运行时间少一半。

A fast codebook training algorithm without the computation of distortion by the property of centroid sequence convergence is presented in this paper. The performance of the new algorithm is tested by typical test image Lena. The result shows that the PSNR difference between the new algorithm and LBG is O.ldB, but the running time of it is at most one half of LBG.