模糊聚类法实现初始码书的优化

应用模糊数学理论,在限定条件下,从训练矢量的空间分布出发,对训练矢量空间进行胞腔划分,通过动态控制λ截集水平,调整截集数量和相应胞腔大小,然后求出相应的形心作为代表矢量构成一个优良的初始码书。采用该初始码书作为LBG矢量量化算法的叠代初值比分裂法、随机法更优。

Realize the Optimization of the Initial Codebook by Using the Fuzzy Clustering Method

Starting from the space distribution of the trained vectors under the limited conditions by using the theory of the fuzzy mathematics, this paper makes the cell partition for the trained vector space through controlling -cut level dynamically and regulating the number of cuts and the size of the corresponding cell, finds the corresponding centers being as the representative vectors and being turned into an excellent initial codebook, and points out that the initial iterative value of the quantitative algorithm adopting the initial codebook as the LBG vector is much better than the disintegrating method and randomizing method.