基于LBP层次特征的表情识别算法

Facial expression recognition algorithm based on feature extraction method of LBP

提出了一种基于LBP层次特征提取的表情识别算法.将图像分成许多子块,并从子块中提取面部运动单元信息来组成基于面部运动单元的表情成分特征,对人脸图像的眼睛和口部作粗定位.采用局部二值模式(LBP)的层次特征提取法,对图像进行分块操作,求出每个子块的LBP直方图,然后将基于整体特征得到的LBP直方图与基于局部特征得到的LBP直方图连接起来,作为整幅图像的LBP直方图.将层次特征提取法所提取的LBP直方图作为嵌入式隐马尔可夫模型(EHMM)的初始向量,即形成观察序列.对JAFFE人脸库中的7种基本表情进行了测试,结果表明该方法能有效提高表情识别率.

：A facial expression recognition algorithm based on the feature extraction method of LBP was intro- duced. Firstly, face images were divided into local patches, and the information of facial action units was ex- tracted to compose the compositional features of facial expressions. The coarse position of eyes and mouth was located. Secondly, the feature extraction method of local binary patterns （LBP） was adopted to decompose the images into sub-blocks, and each sub-blocks histogram was attained. Then the LBP histogram of overall fea- tures and LBP histogram of local features were combined into the LBP histogram of a whole image. Finally, the LBP histogram of a whole image was sent to an embedded hidden Markov model （EHMM） as an initial vector, and the observation sequence was formatted. Through testing seven basic expressions in JAFFE face database, the experiments show that the algorithm can enhance facial expression recognition effectively.