摘要
为了改善矢量量化的码书性能和提高神经网络的学习效率,在分析等误差自组织特征映射算法(equidistortion self-organizing feature mapping,EDSOFM)的基础上,提出了一种改进算法。改进算法将模糊神经网的隶属度函数引入到竞争学习算法中,有效地提高了学习收敛速度。针对原算法搜索获胜码字时计算量较大的问题,改进算法通过不等式判决的方法,快速排除了大量的不匹配码字。实验结果表明,改进算法使码书设计的计算量得到明显的减少,而且码书的性能得到了提高。
In order to promote the codebook＇s performance and the learning eficiency of neural network in the vector quantization,this paper proposes an improved algorithm based on the analysis of equidistortion self-organizing feature mapping algorithm（EDSOFM）.The membership function of fuzzy neural network is introduced into the competiton learning algorithm to converge faster with better performance.For the large computational complexity in searching the winners of ccdbook,the improved algorithm eliminates a large number of unmatched codeword though inequality decision.Simulation shows that the computation is substantially reduced in the codebook design,and the codebook performance is obviously improved.
出处
《重庆邮电大学学报（自然科学版）》
北大核心
2011年第2期155-160,共6页
Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)
基金
国家自然科学基金(61071116)
国家科技重大专项(2009ZX03001-004)
重庆市科委自然科学基金(CSTC
2010BB2407)
信号与信息处理重庆市市级重点实验室建设项目(CSTC
2009CA2003)
重庆邮电大学自然科学基金(A2009-27)~~
关键词
矢量量化
自组织特征映射
等误差原则
模糊神经网
快速搜索
vector quantization
self-organizing feature mapping
equidistortion principle
fuzzy neural network
fast search
