混合样本融合边缘信息的单样本人脸识别研究  

Study on Single Sample Face Recognition Based on Mixed Sample Fusion Edge Information

作　　者：向洋 董林鹭 宋弘 余坤键 XIANG Yang;DONG Lin-lu;SONG Hong;YU Kun-jian(School of Automation and Information Engineering,Sichuan University of Science and Engineering,Zigong 643000,China)

机构地区：[1]四川轻化工大学自动化与信息工程学院,四川自贡643000

出　　处：《计算机技术与发展》2020年第5期66-69,共4页Computer Technology and Development

基　　金：四川省教育项目基金(17ZB0302);四川理工学院科研项目(2015RCA9)。

摘　　要：在进行人脸识别的时候,训练样本数量对识别率的大小影响非常大,由于存储技术和训练样本采集困难等诸多条件的限制,如何利用一幅人脸有用的信息尽可能地生成并包含更多的人脸信息成为了学术界的难点。针对该问题,提出一种按不同权值将原始图像和虚拟样本混合后再融合其人脸不同灰度值的边缘信息,构成新的训练样本。首先将原始样本灰度处理后生成轴对称图像和镜像图像,按不同权值混合。再提取混合后的边缘信息按不同灰度值与混合后的图像融合。使单幅人脸图像包含更多的特征信息。实验结果表明,混合权值之和大于1并且融合其边缘信息后生成的训练样本,相比原始样本信息的人脸识别率能提升2%~12%,表明该方法能有效地提高人脸识别率。When performing face recognition,the number of training samples has a great influence on the size of the recognition rate. Due to the limitations of storage technology and the difficulty in collecting training samples,how to use the useful information of a face to generate and contain as much face information as possible has become a difficult in academia. Aiming at this problem,we propose a kind of edge information that mixes the original image and the virtual sample according to different weights and then fuses the different gray values of the human face to form a new training sample. Firstly,the original sample is processed in gray to generate an axisymmetric image and a mirror image,which are mixed according to different weights. The extracted edge information is fused with the mixed image according to different gray values,which makes a single face image contain more feature information. Experiment shows that the training samples generated after the sum of the blending weights is greater than 1 and the edge information is merged can increase the face recognition rate by 2%~12% compared with the original sample information. It is showed that this method can effectively improve the face recognition rate.

关 键 词：人脸识别 虚拟样本 边缘提取 权值融合 协同表示 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]