山西大同大学学报（自然科学版）
山西大同大學學報（自然科學版）
산서대동대학학보（자연과학판）
Journal of Shanxi Datong University (Natural Science Edition)
2015年
4期
29-32,87
,共5页
矢量量化%SOFM神经网络%初始码书%训练时间
矢量量化%SOFM神經網絡%初始碼書%訓練時間
시량양화%SOFM신경망락%초시마서%훈련시간
vector quantization%self-organizing feature maps%initial codebook%training time
矢量量化作为一种高效的数据压缩技术，在语音和图像的编码、传输中都有广泛的应用，其关键在于码书设计。码书的好坏直接影响语音、图像的编码质量。本文针对经典LBG算法对初始码书敏感及整体训练时间较长这两个缺陷，着重研究SOFM算法在这两方面的性质和特点，结果证实SOFM算法相对于LBG算法训练时间较短，且利用SOFM网络设计的码书受初始码书的影响小，具有很强的自适应性，很好的改善了LBG算法在这两方面的缺陷。
矢量量化作為一種高效的數據壓縮技術，在語音和圖像的編碼、傳輸中都有廣汎的應用，其關鍵在于碼書設計。碼書的好壞直接影響語音、圖像的編碼質量。本文針對經典LBG算法對初始碼書敏感及整體訓練時間較長這兩箇缺陷，著重研究SOFM算法在這兩方麵的性質和特點，結果證實SOFM算法相對于LBG算法訓練時間較短，且利用SOFM網絡設計的碼書受初始碼書的影響小，具有很彊的自適應性，很好的改善瞭LBG算法在這兩方麵的缺陷。
시량양화작위일충고효적수거압축기술，재어음화도상적편마、전수중도유엄범적응용，기관건재우마서설계。마서적호배직접영향어음、도상적편마질량。본문침대경전LBG산법대초시마서민감급정체훈련시간교장저량개결함，착중연구SOFM산법재저량방면적성질화특점，결과증실SOFM산법상대우LBG산법훈련시간교단，차이용SOFM망락설계적마서수초시마서적영향소，구유흔강적자괄응성，흔호적개선료LBG산법재저량방면적결함。
Vector quantization as a highly efficient data compression technology has been widely used in voice and image com?pression coding and transmission. The key problem of VQ is codebook design, because codebook has direct impacts on voice and video encoding quality. There are two serious shortcomings about the classic method LBG algorithm. It is sensitive to the initial codebook and training time is long. To solve these two problems, the text mainly research SOFM algorithm property and point of these two aspects, the result confirms that the codebook designed by SOFM network suffers small impact from the initial code book, and it can self-organized proceed study discipline, and have very strong adaptability. So we can see it well improved LBG algorithm's shortcomings in these two aspects.
