摘要
人脸表情变化有其心理学上的理论基础,几何特征通常能直观地描述表情的变化。一般的表情识别方法直接利用几何特征和面部纹理信息进行特征提取,通常需要用到不同的图像算法。针对细致的表情变化特点,基于特征点计算角度几何特征,作为表情识别的重要依据,直接利用特征点的好处是可以避免一些过多的图像处理算法,在特征点比较稳定的情况下,能够得到较满意的识别效果。在支持向量机上采用不同的核函数对表情分类和识别,同时进行交叉验证。对特征向量进行优化和组合使得所选取的特征向量能最大限度地表现表情的本质。在CK+表情库上测试该方法,得到的正确率为95.16%。
The change of facial expression has its theoretical basis of psychology,and geometric features can usually describe the change of expression intuitively.General expression recognition methods directly use geometric features and facial texture information on the process of feature extraction and usually require different image processing algorithms.Based on the characteristics of meticulous expression changes,the angular geometric features are calculated based on feature points,which is an important basis for expression recognition.The advantage of using feature points directly is that it can avoid many image processing algorithms.When the feature points are stable enough,the recognition effect can be satisfied.Different kernel functions were used to classify and recognize expressions on SVM,and cross validation was carried out.The optimization and combination of feature vectors made the selected feature vectors represent the essence of expression to the greatest extent.Our method is tested on CK+expression database,and the accuracy is 95.16%.
作者
吴珂
周梦莹
李高阳
陈增照
何秀玲
Wu Ke;Zhou Mengying;Li Gaoyang;Chen Zengzhao;He Xiuling(National Engineering Research Center for E-Learning,Central China Normal University,Wuhan 430079,Hubei,China)
出处
《计算机应用与软件》
北大核心
2020年第7期120-124,共5页
Computer Applications and Software
基金
教育部人文社会科学研究规划基金项目(17YJA880030)
华中师范大学中央高校基本科研业务费项目(CCNU15A05012)
中央高校基本科研业务费项目(CCNU19QN028)。