视频序列中基于多尺度时空局部方向角模式直方图映射的表情识别

摘要

Local binary pattern (LBP) is conceptually regarded as non-oriented, so it cannot capture suffi-ciently detailed information. Aiming at the problem, the local orientational pattern (LOP) method is pro-posed. It labels the pixels of the image by comparing two orientational differences at two neighboring pixels and encodes the change of the neighborhood orientational difference. Then LOP is extended to three- di-mensional space, the spatiotemporal local orientational pattern (SLOP) is presented. The features obtained from three orthogonal planes are concatenated into a single vector. Finally the multi-scale SLOP histogram is used as face representation and projected onto locality preserving projection space to obtain lower-dimensional feature. Experimental results on Cohn-Kanade and MMI facial expression databases demonstrate that the proposed method outperforms other existing approaches in recognition rate and recognition speed.%针对局部二元模式在概念上是无方向性的，不能充分捕捉详细信息的问题，提出局部方向角模式(LOP)方法。该方法通过比较2个近邻点上2个方向角的差值来标注图像中的像素点，对邻域内方向角差值的变化进行编码；将LOP 扩展到三维空间，提出时空局部方向角模式(SLOP)，将从3个正交平面上提取到的特征串接成一个向量；最后采用多尺度SLOP直方图作为人脸表征，并将其投影到保局映射空间以获取低维特征。在Cohn-Kanade与 MMI人脸表情数据库上的实验结果表明，文中方法在识别准确率和识别速度方面都优于已有方法。