面向遮挡条件下的人脸识别方法研究

项目名称： 面向遮挡条件下的人脸识别方法研究

项目编号： No.61502067

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 其他

项目作者： 栾晓

作者单位： 重庆邮电大学

项目金额： 20万元

中文摘要： 光照、姿态以及遮挡是影响人脸识别系统性能的主要因素，目前尚未得到很好解决。其中，遮挡会造成人脸图像信息缺失，干扰对面部特征的抽取，甚至导致人脸识别算法失效。如何提取不受遮挡影响的人脸特征以及对遮挡人脸进行有效表示是该方向研究重点和难点。本项目将重点研究遮挡条件下人脸识别的关键技术，其目的是进一步提高人脸识别系统的精度和鲁棒性。通过分析遮挡区域对人脸识别算法的影响，本项目的主要研究思路是：研究利用测试图像（带遮挡人脸图像）在不同类别下分解得到的稀疏残差成分抽取人脸的判别性特征；研究将人脸样本向低秩空间投影，同时约束不同类别人脸结构之间的相关性，从而抽取出不受遮挡影响的人脸本质特征；研究在线性回归分类方法中引入权重表示函数，自适应判别测试图像中的遮挡像素点并赋予其较小的权重，以减小其对测试图像编码的影响。本项目的研究有望丰富人脸识别的相关理论和算法，为实际应用提供理论和技术支持。

中文关键词： 模式识别；人脸图像识别；遮挡人脸识别

英文摘要： Illumination, pose and occlusions are major factors affecting the performance of face recognition system, and have not yet been solved. In particular, occlusion can cause the loss of facial information, the interference for facial feature extraction and recognition, and even leads to the failure of face recognition algorithms. How to extract facial features not affected by occlusion and effectively represent the occluded face is the key to solving this problem. This proposal will focus on the key technology of face recognition under occlusion, and aim to further improve the accuracy and robustness of face recognition systems. By analyzing the impact of occlusions on face recognition algorithms, the main idea includes the following aspects. Firstly, we propose to extract discriminative information of sparse residuals of test sample (occluded face image) under different subjects. Secondly, we propose to proposal face samples onto the associated low-rank space, while impose a constraint on correlation between different classes of face structure, and extract the intrinsic features which is robust to occlusion. Finally, we introduce a weight function into the linear regression classification method. The proposed method can adaptively determines the corrupted pixels and assigns small weight on those pixels, and can reduce the impact of occlusions on coding. This research is expected to enrich the theory and algorithms of face recognition, and to provide theoretical and technical supports for the application.

英文关键词： pattern recognition;face recognition;occluded face recognition