基于面部块运动历史直方图特征的视频表情自动识别

AUTOMATIC FACIAL EXPRESSION RECOGNITION BASED ON MOTION HISTORY HISTOGRAM FEATURES OF FACIAL SALIENCY BLOCKS

为了自动识别视频中的表情类别,提出基于面部块表情特征编码的视频表情识别方法。检测并精确定位视频中人脸关键点位置,之后以检测到的关键点为中心,提取面部显著特征块;对面部各特征块提取运动历史直方图动态特征描述子,这些描述子被当作表情特征输入Adaboost分类器进行训练和识别;最终预测视频表情类型。通过在国际通用表情数据库BU-4DFE的纹理图像上进行测试,取得了83.2%的平均识别率,充分证明了所提算法的有效性。跟同领域其他主流算法相比,所提算法具有很强的竞争性。

In order to automatically identify the expression category in the video, we proposed a tully automatic video FER framework. Firstly, the location of the key points of a human face in the video was detected and precisely located, and then the significant feature block was extracted with the key points detected as the center. Secondly, we extracted motion history histograms and feature descriptor from each feature block. These dynamic expression descriptors were input into Adaboost classifier to train and predict the expression type finally. We carried out experiments on BU-4DFE dataset and got a state-of-art 83. 2% average performance which indicates the validity of the proposed approach. Compared with other mainstream algorithms in the same field, the proposed algorithm is highly competitive.