基于FWT-Projection-BPNN的表情识别  

Research on expression-based fast wavelet transform-projection-BP neural network approaches

作　　者：吕英华[1] 崔景霞[1] 

机构地区：[1]东北师范大学计算机学院,吉林长春130117

出　　处：《东北师大学报（自然科学版）》2009年第2期84-91,共8页Journal of Northeast Normal University(Natural Science Edition)

基　　金：吉林省科技发展计划项目(20070322)

摘　　要：提出了一种基于快速小波变换-投影-BP神经网络(FWT-Project-BPNN)的人脸表情识别方法.该算法首先利用快速小波变换(FWT)对表情图像进行变换,以期在不明显损失图像信息的基础上达到压缩数据量的目的.然后分别对变换后的水平方向与垂直方向的高频数据子图做水平方向与垂直方向投影.将得到的水平与垂直向量组成该表情识别算法的特征向量,最后建立一对一的BP神经网络来进行训练.实验结果表明,该算法能够在一定条件下快速且较准确地识别出悲伤、愤怒、高兴、惊讶、恐惧、厌恶、中性7种通用样本表情.This paper presents a method for facial expression recognition based fast wavelet transform- Projection-BP neural network （FWT-Project-BPNN） approaches. Firstly, the fast wavelet transform was carried out to compress the preprocessed images ＇ data on the basis of not losing essential image information. Then after-FWT the horizontal high frequency sub-image was projected to a horizontal vector and the vertical high frequency sub image to a vertical vector respectively. The feature vectors were got by connecting the expression images ＇ horizontal projection with transposed vertical projection. Finally,BP neural network （BPNN） is used to make a classification. Experimental results indicate that the proposed method can be simple, high ration of recognition, but it still need to be improved on the capability of extensiveness.

关 键 词：人脸表情识别 快速小波变换 BP神经网络 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]