模糊人脸同步恢复与识别
发布时间：2018-11-16 21:21
【摘要】：模糊是一种最常见的图像退化之一。对于模糊的人脸,通常面临着两大任务:模糊人脸识别和模糊人脸恢复。目前的大量工作都只针对其中某一个任务,只有极少量的工作将两个任务一起考虑。本文分析研究了模糊人脸识别和恢复之间的相辅相成关系。对于模糊人脸恢复,提出了两个新的基于模糊人脸识别结果的模型,分别解决了当前最好的基于样例人脸去模糊方法的两个缺陷。模型1利用训练类内差词典的线性表示解决了人脸类内差距问题,模型2利用梯度的L0.8先验替代L0先验解决了清晰的纯人脸区域先验约束问题。两个模型级联成两步方案解决基于人脸识别的模糊人脸恢复问题。对于模糊人脸识别,综合对比了模糊和去模糊方法,选择利用模糊方法和LPQ特征做人脸识别解决基于模糊人脸恢复的模糊人脸识别问题。最后,将恢复与识别构成一个良性循环,提出了一个模糊人脸同步恢复与识别(Simultaneous Blurred Face Restoration and Recognition,SRR)算法,迭代完成模糊人脸恢复与识别。本文提出的SRR算法适用于复杂模糊核情况下的人脸去模糊问题,还适用于小样本情况下的模糊人脸识别问题。在FERET数据库上的实验表明,对于多种不同的模糊,SRR算法不仅大幅提高了模糊人脸识别的准确率,还大大提升了模糊人脸恢复的质量。
[Abstract]:Blur is one of the most common image degradation. For fuzzy faces, there are usually two tasks: fuzzy face recognition and fuzzy face restoration. Much of the current work is focused on only one of these tasks, with very little work considering the two tasks together. This paper analyzes and studies the complementary relationship between fuzzy face recognition and restoration. For fuzzy face restoration, two new models based on fuzzy face recognition results are proposed, respectively, to solve the two defects of the best face de-blurring method based on sample examples. Model 1 uses the linear representation of the training intra-class difference dictionary to solve the intra-class gap problem. Model 2 uses gradient L0.8 priori to replace L0 priori to solve a clear priori constraint problem in pure face regions. The two models are cascaded into two steps to solve the fuzzy face restoration problem based on face recognition. For fuzzy face recognition, fuzzy and de-fuzzy methods are compared synthetically, and fuzzy face recognition problem based on fuzzy face restoration is solved by using fuzzy method and LPQ feature as face recognition method. Finally, a fuzzy face synchronous recovery and recognition (Simultaneous Blurred Face Restoration and Recognition,SRR) algorithm is proposed, which is composed of a benign cycle of restoration and recognition, and the fuzzy face recovery and recognition is completed iteratively. The SRR algorithm proposed in this paper is suitable for the human face deblurring problem in the case of complex fuzzy kernel and for the fuzzy face recognition problem in the case of small samples. Experiments on FERET database show that SRR algorithm not only improves the accuracy of fuzzy face recognition, but also improves the quality of fuzzy face restoration.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41
[Abstract]:Blur is one of the most common image degradation. For fuzzy faces, there are usually two tasks: fuzzy face recognition and fuzzy face restoration. Much of the current work is focused on only one of these tasks, with very little work considering the two tasks together. This paper analyzes and studies the complementary relationship between fuzzy face recognition and restoration. For fuzzy face restoration, two new models based on fuzzy face recognition results are proposed, respectively, to solve the two defects of the best face de-blurring method based on sample examples. Model 1 uses the linear representation of the training intra-class difference dictionary to solve the intra-class gap problem. Model 2 uses gradient L0.8 priori to replace L0 priori to solve a clear priori constraint problem in pure face regions. The two models are cascaded into two steps to solve the fuzzy face restoration problem based on face recognition. For fuzzy face recognition, fuzzy and de-fuzzy methods are compared synthetically, and fuzzy face recognition problem based on fuzzy face restoration is solved by using fuzzy method and LPQ feature as face recognition method. Finally, a fuzzy face synchronous recovery and recognition (Simultaneous Blurred Face Restoration and Recognition,SRR) algorithm is proposed, which is composed of a benign cycle of restoration and recognition, and the fuzzy face recovery and recognition is completed iteratively. The SRR algorithm proposed in this paper is suitable for the human face deblurring problem in the case of complex fuzzy kernel and for the fuzzy face recognition problem in the case of small samples. Experiments on FERET database show that SRR algorithm not only improves the accuracy of fuzzy face recognition, but also improves the quality of fuzzy face restoration.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41
【相似文献】
相关期刊论文 前10条
1 齐亚莉;李业丽;;模糊/去模糊攻击的研究[J];北京印刷学院学报;2005年04期
2 姚敏;去模糊机制及其一般形式[J];自动化仪表;1997年12期
3 徐静;石玉英;王肖玉;;全变分图像去噪和去模糊问题的外插边值条件(英文)[J];计算物理;2007年03期
4 马丽红,张宇;波动方程去模糊的累积噪声的CURVELET变换消除方法[J];信号处理;2005年05期
5 王国栋;潘振宽;刘存良;郑世秀;;基于混合数据项的运动去模糊变分方法[J];仪器仪表学报;2013年07期
6 唐梦;彭国华;郑红婵;;基于正则化方法的图像盲去模糊[J];计算机应用研究;2014年02期
7 徐志浩;靳东明;李志坚;;基于电荷泵实现的模糊控制器去模糊电路设计[J];微电子学;2005年06期
8 裴玉龙;仝卫国;李宝树;;航拍绝缘子图像的去模糊处理[J];电瓷避雷器;2010年05期
9 唐敏;彭国华;;基于非均匀模糊核的RL改进算法[J];计算机工程与应用;2013年21期
10 孙英皓;唐棣;;一种有效的交互式图像局部内容放大方法[J];计算机工程与应用;2012年12期
相关会议论文 前1条
1 张一鸣;刘亚t，
本文编号：2336673
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2336673.html
