Computer Engineering and Applications ›› 2019, Vol. 55 ›› Issue (14): 134-141.DOI: 10.3778/j.issn.1002-8331.1804-0118

Previous Articles     Next Articles

Face Recognition Based on SA-CRC of Multi-Layer AR-LBP and WLD Feature Fusion

YE Feng, YE Xueyi, LUO Xiaohan, CHEN Ze   

  1. Lab of Pattern Recognition & Information Security, Hangzhou Dianzi University, Hangzhou 310018, China
  • Online:2019-07-15 Published:2019-07-11


叶  枫,叶学义,罗宵晗,陈  泽   

  1. 杭州电子科技大学 模式识别与信息安全实验室,杭州 310018

Abstract: In view of the limitation of the facial feature fetched by Asymmetric Region Local Binary Pattern(AR-LBP) as well as the interclass interference of faces in the Collaborative Representation based Classification(CRC), a supplementary joining the features of multi-layer AR-LBP and features of Weber Local Descriptor(WLD) is proposed, and the interclass interference is reduced by augmenting the sparsity in CRC. Firstly, the multi-layer AR-LBP features of face images are extracted and cascaded. Then AR-LBP features are integrated with the WLD features extracted from the original image, so the multi-layer AR-LBP and WLD fusion features are obtained. Finally, the Sparsity Augmented Collaborative Representation based Classification(SA-CRC) is used to complete the classification of faces. In the face database of ORL, Yale and GT, the recognition accuracy of fusion features in this paper is increased by 0.7%~42.6% compared with AR-LBP, WLD, and multi-layer LBP and HOG fusion features. When CRC is replaced with SA-CRC, the recognition accuracy is further improved.

Key words: Asymmetric Region Local Binary Pattern(AR-LBP), Weber Local Descriptor(WLD), Collaborative Representation based Classification(CRC), Sparsity Augmented Collaborative Representation based Classification(SA-CRC), feature extraction

摘要: 针对非对称局部二值模式(AR-LBP)提取的人脸特征有限,以及协同表示分类(CRC)人脸存在的类间干扰,提出以多层AR-LBP特征及联合韦伯局部描述子(WLD)特征进行补充,并以增加CRC中稀疏性来降低类间干扰。提取人脸图像的多层AR-LBP特征并级联,与从原图像提取的WLD特征级联得到多层AR-LBP与WLD融合特征,采用稀疏增强的协同表示分类(SA-CRC)完成人脸分类。在ORL、Yale和GT公开人脸库上,提出的多层AR-LBP与WLD特征融合算法与AR-LBP特征提取算法、WLD特征提取算法以及多层LBP与HOG特征融合算法相比,识别正确率提高了0.7%~42.6%;当利用SA-CRC取代CRC后,识别正确率进一步得到提高。

关键词: 非对称局部二值模式(AR-LBP), 韦伯局部描述子(WLD), 协同表示分类(CRC), 稀疏增强的协同表示分类(SA-CRC), 特征提取