基于HOG低秩恢复与协同表征的人脸识别方法

在获取到的人脸图像不完备以及人脸图像在有遮挡、光照、表情的变化或受到噪声污染时,识别率就会变得十分低,针对这一问题,本文提出了一种基于HOG低秩恢复与协同表征的人脸识别算法HLRR_CRC.首先采用低秩恢复算法得到训练样本和测试样本的干净人脸图像,然后对测试样本中干净的人脸图像和训练样本中干净的人脸图像分别进行HOG特征提取,得到HOG特征向量,以此特征向量为基础,得到测试样本特征矢量的协同表示,最后,通过规则化残差进行分类.在ORL、Extended Yale B和AR数据库上进行测试,实验结果表明,本文算法对光照、噪声较鲁棒,相比于当前的人脸识别算法,本文算法在恶劣光照和噪声下的识别率平均提高29.6％.

Face image recognition method via HOG low-rank recovery and collaborative representation

It has been a key problem in face recognition which the recognition rate tend to be low when acquired face images are incompleted or face images are polluted by noise or in sheltered,illumination and expression changes.Thus,a face recognition algorithm based on low-rank recovery and cooperative representation (HLRR_CRC) is proposed.First,clean images of train samples and test samples are obtained by LRR algorithm.Then,HOG feature vectors are obtained from clean face images by HOG feature extraction,and the test sample feature vector is cooperative represented based on train sample vectors.Finally,the test samples are classified by reconstruction error.Test on ORL,Extended Yale B and AR database show that the algorithm is robust to illumination and noise.Under harsh illumination and noise conditions,the average recognition rate of the algorithm is improved by 29.6％ compared to the current face recognition algorithm.