摘要
针对人脸表情识别背景复杂导致识别率低的缺点,提出了一种中心对称三值模式(CSTP)算法,首先对人脸表情图像进行分块处理,在每一子块上提取CSTP特征,并对该子块进行CSTP特征的直方图统计,然后求出各个子块对应的信息熵,构造自适应加权系数,再分别和各个子块的直方图相乘,将自适应加权后的各个子块特征向量级联作为最终的纹理特征,最后利用支持向量机(SVM)进行表情分类。在JAFFE和CMU-AMP表情库上进行试验,通过对比其他传统方法发现该算法对表情识别更有效。
To solve the disadvantage of low recognition rate which caused by complex facial expression background, a new algorithm called Center Symmetrical Ternary Patterns (CSTP) is proposed in this paper. Firstly, the facial expression image is divided into blocks, CSTP features are extracted from each sub-block and the histogram statistics of the CSTP feature of the sub-block is car- ried out. Then the information entropy of the each sub-block has been calculated in order to construct the adaptive weighted coefficient. The feature vectors of the adaptive weighted sub-block have been scaled up as the final texture features. Finally, Support Vector Machine (SVM) is used to expression classification. The experiments on JAFFE and CMU-AMP database have been compared with other traditional methods. The better results of the proposed algorithm have been achieved in expression classification.
出处
《电视技术》
北大核心
2016年第2期127-131,共5页
Video Engineering
基金
河北省高等学校自然科学研究重点基金项目(ZD20131043)
关键词
表情识别
自适应加权
支持向量机
expression recognition
adaptive weighted
support vector machine