误差敏感竞争性学习算法

A Distortion Sensitive Competitive Learning Algorithm for Vector Quantization

本文基于等误差准则提出了一种适用于矢量量化技术的新型码书设计算法。实验表明此算法优于现存算法。为解决初始码书赋值问题，本文提出了自生成自组织神经网络方法。实验表明此算法加速了算法的收敛速度。

In this paper, a new competitive learning algorithm based on equidistortion principle for vector quantization (VQ) design was proposed. The experimental results showed that the new algorithm has a better performance than the other known algorithms. To solve the problem of initializing codebooks for VQ, the self creating and organizing neural network (SCONN) was introduced. The experimental results showed that the SCONN algorithm had improved the performance of the known learning algorithms and made them converge faster.