神经网络矢量量化的设计与实现

Design and implementation of vector quantization based upon a hybrid neural network

设计了一种混合神经网络矢量量化编码方法。利用Ｋｏｈｏｎｅｎ网络的自组织聚类功能设计矢量量化器码书，实现矢量量化，用３层ＢＰ网络完成码字的信道符号编码。该神经网络矢量量化编码器能够并行处理输入矢量，速度快，效率高。

A coding scheme for vector quantization is presented by using a hybrid neural network. The clustering function of the Kohonen self organizing feature map is used to design codebook of vector quantizer and to carry out vector quantization, then the channel index for codeword is made by a three layers BP (Back Propagation) network. Processing input vectors in parallel mode fast and effectively, the vector quantization coder based upon neural networks is sutable for voice and image data compression.