基于熵加权Gabor特征的鲁棒人脸识别方法

A Robust Face Recognition Method Based on Entropy Weighting Gabor Features

针对传统人脸识别方法对于人脸姿态变化中偏转人脸的识别率较低的问题,提出一种基于监督下降法(SDM)模型和熵加权Gabor特征的鲁棒人脸识别方案。首先,利用SDM模型提取人脸图像局部特征,对人脸进行矫正对齐;然后,将人脸构建成一个变形网格,并计算Gabor特征系数(Gabor jets);再后,引入图像的熵来加权Gabor jets特征,并进行局部归一化(LN);最后,利用Borda计数将输入人脸图像上的Gabor jets与图库进行比较,并通过统计模型对人脸进行分类识别。实验结果表明,该方案能够很好的识别具有姿势、光照和表情变化的人脸图像,特别对人脸偏转具有很强的鲁棒性。

For the issues that the traditional face recognition method has lower recognition rate of rotated face in face pose variation, this paper develops a robust face recognition scheme based on the supervised descent method and entropy weighting Gabor features. Firstly, it uses the supervised descent method to extract local features of face images and to align the face. Then, this scheme uses the entropy of the image to weight the Gabor jets features, and execute local normalization （LN）. Finally, it uses Borda count to compare the Gabor jets of input facial images with the library image, and uses the statistical model to realize face recognition. Experimental results show that the proposed scheme can recognize face images with pose, illumination and facial expression changes, especially for face deflection.