基于小波MCBP和WEF的人脸表情识别  

Facial expression recognition based on wavelet transformed MCBP and WEF

作　　者：胡敏[1] 陈杏[1] 王晓华[1] 许良凤[1] 李瑞[1] 

机构地区：[1]合肥工业大学计算机与信息学院,合肥230009

出　　处：《电子测量与仪器学报》2012年第11期927-932,共6页Journal of Electronic Measurement and Instrumentation

基　　金：国家自然科学基金-广东联合基金重点项目(编号:U1135003)资助项目;国家863高技术研究发展计划资助项目(编号:2012AA011103)资助项目

摘　　要：现有的多尺度中心化二值模式(MCBP)通过在原始图像上改变CBP算子的半径,随着算子半径的增加计算量也迅速增加。针对这个问题,提出一种基于小波的MCBP(WMCBP)的人脸表情识别方法,对小波分解后的两幅低频图像的特征区域进行CBP变换,得到多级局部CBP直方图序列特征.该方法不仅能获得更加准确的多尺度信息,而且大大降低了运算量。为进一步提高表情识别率,引入了加权的小波能量特征(WWEF)。通过对JAFFE人脸表情库的实验证明;这两部分特征在一定程度上可互补,将它们融合能在不明显增加运算量的前提下增强WMCBP的表情识别能力。The existing multi-scale centralized binary pattems get multi-scale features by changing the radius of CBP operator on the original image, with the increase of the radius of operator, the computation of the algorithm increase rapidly. To deal with this problem, a facial expression recognition method based on Wavelet Transformed MCBP is developed in this paper. The method can not only get more accurate multi-scale information, but also greatly reduces the computation complexity.Furthermore, WWEF is introduced in order to enhance the facial expression recognition accuracy. Experiments on JAFFE facial expression database show that these two types of features are complementary to some extent, and the fusion of them can enhance the performance of WMCBP in facial expression recognition without increasing the computation obviously.

关 键 词：表情识别 多尺度中心化二值模式 小波 小波能量特征 特征融合 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]