基于面部显著块动态信息的视频表情自动识别

摘要

To predict the expression type automatically from the video sequence,a fully automatic video FER framework was proposed.The key points of human face in video were detected using regressing local binary features (LBF) algorithm.The areas with a certain width around the detected points were segmented and the corresponding LBP-T(OP (local binary pattern from three orthogonal planes) features along the video sequence were calculated.These dynamic expression descriptors were fed into Adaboost classifier to train and predict the expression type.Experiments were carried out on BU 4DFE dataset.A 81.2％ average performance is got which indicates the validity of the proposed approach.%为自动识别视频中表情类别,提出基于面部块表情特征编码的视频表情识别方法框架.检测并精确定位视频中人脸关键点位置,以检测到的关键点为中心,提取面部显著特征块.沿着时间轴方向,对面部各特征块提取LBP-T(OP(local binary pattern from three orthogonal planes)动态特征描述子,将这些描述子作为表情特征并输入Adaboost分类器进行训练和识别,预测视频表情类型.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81.2％的平均识别率,验证了所提算法的有效性,与同领域其它主流算法相比,其具有很强的竞争性.