在矢量量化中,码书起决定性的作用,它决定了量化的性能;一般采用LBG算法生成码书,其中一个关键的技术就是初始码书的选取,通常认为分裂法效果显著.该文引入贪婪树生长算法来设计初始码书,得到了两个优于分裂法的初始码书设计算法,减少了整个码书训练的运算时间,提高了码书的性能.

Codebooks are crucial to vector quantizations (VQs), which determine their performance. LBG algorithm is often used to generate codebooks, where choice of original codebooks is a key technique, and the splitting algorithm is generally considered effective. In this paper, the greedy tree growing algorithm is introduced to design original codebooks, and two better algorithms are derived. Compared to the classical splitting algorithm, the total run time with the proposed algerithm is reduced, and the codebook performance is improved.