摘要
针对传统的稀疏表示分类算法中面部对齐受限而影响人脸识别率的问题,提出一种基于约束采样和面部对齐的稀疏表示分类算法。首先通过使用约束采样对训练图像进行预先标注得到固定脸特征;然后结合图像的纹理信息和形状特征进行面部对齐及特征提取;最后计算出测试样本与各个训练样本之间的相似度,利用稀疏表示分类器完成人脸的识别。在AR、CAS-PEAL及扩展YaleB人脸数据库上的实验验证了算法的有效性及鲁棒性。实验结果表明,约束采样和面部对齐的组合大大提高了人脸识别率,相比几种较为先进的鲁棒人脸识别算法,该算法取得了更好的识别效果。
For the issue that face alignment is limited in traditional sparse representation classification method which will impact face recognition rate, we propose a sparse representation classification method, it is based on constraint sampling and face alignment. First, we mark the training images in advance by using constraint sampling to get fixed face features. Then we do the face alignment and extract the features in combination with image texture information and shape features. Finally, we calculate the similarities between testing image and each training image and use sparse representation classifier to complete the face recognition. The effectiveness and robustness of the proposed algorithm are verified by the experiments on face databases AR, CAS-PEAL and extended YaleB. Experimental results show that the combination of constraint sampling and face alignment greatly improves face recognition rate. This algorithm achieves better recognition effect than several other advanced robust face recognition algorithms.
出处
《计算机应用与软件》
CSCD
北大核心
2014年第10期192-196,共5页
Computer Applications and Software
关键词
人脸识别
稀疏表示分类
约束采样
面部对齐
光照变化
面部伪装
Face recognition
Sparse representation classification
Constraint sampling
Face alignment
Illustration variation
Facial masking
