结合D-SIFT和区域PPWLD的人脸表情识别方法  

FACIAL EXPRESSION RECOGNITION METHOD COMBINED WITH D-SIFT AND REGIONAL PPWLD

作　　者：付思亚 胡西川[1] Fu Siya;Hu Xichuan(College of Information Engineering,Shanghai Maritime University,Shanghai 201306,China)

机构地区：[1]上海海事大学信息工程学院,上海201306

出　　处：《计算机应用与软件》2022年第7期207-214,共8页Computer Applications and Software

摘　　要：SIFT特征提取方法所提取出的关键点数目及位置不确定、针对性不足,WLD特征仅关注周围像素差异,不能充分反映图像空间结构信息。对此提出结合D-SIFT和PPWLD的人脸表情识别方法。使用D-SIFT,采用级联回归树进行人脸关键点检测,筛选出与表情密切相关的特征点,使用SIFT描述方法进行描述;同时使用PPWLD,利用Prewitt算子改进原始WLD方法的差分激励算子和方向算子,针对关键表情区域进行特征描述。将得到的两种特征融合使用支持向量机进行分类。结果表明,该方法的识别率在人脸表情数据集JAFFE上达到了97.39%,CK+上达到了99.35%,证明其能够有效提高识别率。The number and location of key points extracted by SIFT feature extraction are uncertain and insufficiently targeted. And the WLD feature only focuses on the differences of surrounding pixels, which cannot fully reflect the image spatial structure information. Therefore, we propose a facial expression recognition method combined with D-SIFT and PPWLD. During using D-SIFT, the cascade regression tree was used to detect the face key point and to select feature points closely related to expressions. And SIFT description method was adopted to describe. The Prewitt operator was applied to improve the differential excitation operator and direction operator of the original WLD method. The features were described for key expression regions. We collaborated the two features and used SVM to classify. The experimental results show that the recognition rate reaches 97.39% on face expression dataset named JAFFE and 99.35% on CK+. It is verified that this method can effectively improve the recognition rate.

关 键 词：PPWLD D-SIFT PREWITT算子 直方图 表情识别 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]