摘要
基于压缩感知技术,提出了一种面向遮挡的人脸识别算法。首先,将图像分成各个局部小块,并构建相似遮挡区域;然后,重构图像碎片,从而检测遮挡区域;最后,利用非遮挡区域获取遮挡截面,投票机制完成人脸识别。实验结果显示,其算法在AR和LFW人脸库上的最高识别率分别可高达99.8%和83.8%,优于其他几种遮挡人脸识别算法,此外,该算法对不同遮挡级别的人脸具有较好的鲁棒性。
A face recognition algorithm based on compressed sensing technology for occlusion is proposed.Firstly,images are divided into each local small piece,and similar occlusion areas are constructed.Then,images fragments are reconstructed to detect occlusion area.Finally,non-occlusion areas are used to get occlusion section,and voting mechanism is used to finish face recognition.Experimental results show that the face recognition accuracy of proposed algorithm can achieve at 99.8%on AR and 83.8%on LFW databases,respectively,which is higher than several other occlusion algorithms.Besides,proposed algorithm has good robustness for face with different occlusion levels.
出处
《微型电脑应用》
2016年第4期36-39,共4页
Microcomputer Applications
基金
新疆维吾尔自治区自然科学基金项目(2013211A031)
关键词
人脸识别
压缩感知
遮挡区域检测
图像碎片重构
鲁棒性
投票机制
Face Recognition
Compressed Sensing
Occlusion Area Detection
Image Fragment Reconstruction
Robustness
Voting
