基于Log-Gabor和ULBP改进算法的人脸识别

为进一步提升人脸识别系统的识别率,加强其对光照、表情、姿态变化的鲁棒性,针对人脸识别中的特征提取问题,提出一种基于Log-Gabor与均匀局部二值模式（Uniform Local Binary Pattern,ULBP）改进算法的人脸识别方法。该算法采用多尺度、多方向Log-Gabor滤波器对图像进行滤波来提取Log-Gabor特征,再通过旋转不变均匀模式的LBP进行运算编码,并利用局部空间直方图来描述人脸,最后通过加权的卡方距离对直方图匹配完成人脸识别。在Yale、GT人脸数据库上的测试结果表明,该方法具有更好识别性能,且对环境鲁棒性较好。

Face recognition based on Log-Gabor and ULBP improved algorithm

A face recognition method based on the Log-Gabor and the Uniform Local Binary Pattern improved algorithm is proposed according to the problems of feature extraction in face recognition. This method can further improve the recognition rate of face recognition system, and enhance the robustness to changes of illumination, facial expression and posture. In this algorithm, the multi-scale, multi-direction Log-Gabor image filter is first used to extract Log-Gabor features, and then the operation and encoding are carried out using the rotation invariant uniform pattern of LBP. Next, the local spatial histogram is used to describe the face. Finally, face recognition is completed by matches through the chi-square distance weighted histogram. Experimental tests are carried out on Yale and GT face database andtest results show that the proposed method has better recognition performance and robustness to the environment.