基于Gabor小波变换的人脸表情识别

Facial Expression Recognition based on Gabor Wavelet Transform

在人脸表情识别中,针对Gabor小波变换特征维数很大的问题,提出了一种新的多方向特征编码方法。通过对Gabor特征幅值进行统计处理,将每个像素点同一尺度不同方向的Gabor特征幅值阈值化成二进制,加强了Gabor小波对图像局部结构信息的表征。同时,结合了类似旋转不变LBP的方法对图像进行降维。为了进一步提高表情的正确识别率,采用一种局部区域融合的方法,最后在JAFFE表情库上进行测试,得到比较好的识别率,验证了所提方法的有效性。

Aiming at the high dimensions of Gabor wavelets feature in facial expression recognition, a new method for coding Gabor multi-orientation features is proposed. After statistical processing of the Gabor features and binary processing of the magnitudes of different orientations, the local structural information representation of the image by Gabor wavelet is enhanced. Meanwhile, for reducing the image dimensions, a similar method as in the Rotation Invariant LBP is used and discussed. Furthermore, a method for fusing the local facial regions is employed, thus to improve the expression recognition rate. The experiment on JAFFE database indicates a fairly good recognition rate, and this verifies the effectiveness of the proposed algorithm.