局部遮挡条件下的人脸表情识别  

FACIAL EXPRESSION RECOGNITION UNDER PARTIAL OCCLUSION

作　　者：李蕊[1] 刘鹏宇[1] 贾克斌[1] 

机构地区：[1]北京工业大学电子信息与控制工程学院,北京100124

出　　处：《计算机应用与软件》2016年第9期147-150,175,共5页Computer Applications and Software

基　　金：国家科技支撑计划项目(2011BAC12B03);北京市博士后工作经费项目

摘　　要：针对局部遮挡条件下的人脸表情识别,提出一种新的基于Gabor滤波和灰度共生矩阵的表情识别算法。首先设计一种分块提取Gabor特征统计量的方法,生成一个低维Gabor特征向量;然后,考虑到分块的Gabor特征缺失了像素之间的关联性,将反映像素间位置分布特性的灰度共生矩阵引入到表情识别领域,以此来弥补Gabor特征分块处理产生的不足;最后,将提取的低维Gabor特征向量和灰度共生矩阵纹理特征进行线性叠加,高斯归一化后生成一组用于特征表达的低维特征向量。在日本女性人脸表情库和荷兰内梅亨大学人脸数据库上的实验证明该算法对人脸不同区域、不同程度遮挡的表情识别具有鲁棒性强、特征向量维数低、分类耗时短、识别速率高的特点。We propose a novel facial expression recognition method, which is based on Gabor filter and gray-level co-occurrence matrix, aimed at facial expression recognition under partial occlusion condition. We first design an approach to extract in blocks the Gabor feature statistics, which generates a low-dimensional Gabor feature vector. Then, taking into account the lack of association between pixels in blocked Gabor features, we introduce the gray-level co-occurrence matrix reflecting the distribution characteristics between locations of pixels into expression recognition field, so as to make up the deficiency caused by Gabor feature blocking processing. Finally, we apply the linear superimposition on the extracted low-dimensional Gabor feature vector and the texture feature of gray-level co-occurrence matrix, after Gaussian normalisation processing there generates a set of low-dimensional feature vectors for feature representation. Experiments have been made on JAFFE and RaFD, they prove that the algorithm has the characteristics of high robustness, low dimension of feature vectors, short classification time and better recognition rates on facial expression recognition in different regions and with different occlusion degrees.

关 键 词：人脸表情识别 局部遮挡 Gabor滤波灰度共生矩阵 高斯归一化 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]