基于视觉机制与协同显著性的自发式表情识别  

Spontaneous expression recognition based on visual mechanism and synergetic significance

作　　者：姬秋敏 张灵[1] 陈云华[1] 麦应潮 向文 罗源[1] JI Qiu-min;ZHANG Ling;CHEN Yun-hua;MAI Ying-chao;XIANG Wen;LUO Yuan(School of Computers,Guangdong University of Technology,Guangzhou 510006,China)

机构地区：[1]广东工业大学计算机学院

出　　处：《计算机工程与设计》2019年第6期1741-1746,共6页Computer Engineering and Design

基　　金：广东省交通运输厅科技基金项目(科技-2016-02-030);广东省自然科学基金博士启动基金项目(2014A030310169)广东省自然科学基金面上基金项目(2016A030313703)广东省自然科学基金项目(2016A030313713);广东省科技计划基金项目(2016B030305002)

摘　　要：自发性面部表情视频中存在过多的冗余图像数据而在性能上遇到瓶颈,为此提出一种基于人眼视觉注意机制和协同显著性人脸特征提取算法。基于人类的视觉机制,动态提取关键帧解决面部表情的时间动态问题。通过乘法融合算法和图像协同获得关键序列帧显著区域。由于这些区域形变较小且与面部表情低相关,剔除这些显著区域,减少面部特征数据量。提取剩余面部的LBP特征,并使用SVM分类器进行分类。在Cohn-Kanade+人脸表情数据库和MMI数据库中的实验结果表明,该方法能有效提高自发表达序列的识别准确率。Most relative algorithms encounter bottle-necks in performance due to big redundant images data in the video. A co-salient facial feature extraction algorithm was proposed, based on human visual attention mechanism and group data co-processing technology. Based on human visual mechanism, key frames of expression were dynamically derived from sampling videos to solve the temporal dynamics of facial expressions. Using key sequence frames, salient regions were obtained using multiplicative fusion algorithm in multi-images co-operative manner. These salient regions were got rid of due to their little deformation and low-correlation to facial expressions, and the amount of facial features data was reduced. LBP features of the remainder of facial features were extracted and SVM classifier was used to classify them, respectively. In Cohn-Kanade+facial expression database and MMI database, experimental results show that the proposed method can effectively improve the recognizing accuracy of spontaneous expression sequence.

关 键 词：动态连续自发式 动态采样 协同显著性 视觉注意机制 分类 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]