基于压缩感知的人脸识别系统设计及实现

【作者】 张涛；

【导师】 吴键；

【作者基本信息】 南京理工大学 ， 精密仪器与机械， 2017， 硕士

【摘要】 随着科技的发展,传统的身份识别系统的安全性是远远不够的,而基于生物特征的人脸识别技术是构建新的身份识别系统的热点和难点。压缩感知(CS)是一种新的采样理论,近年来,压缩感知与人脸识别技术的融合是图像识别领域的新的热点。本文对稀疏表示分类算法(SRC)做了相关研究与仿真,并针对SRC算法的一些不足做了一些改良,构建了一个与压缩感知理论结合的人脸识别系统。本文的研究工作如下所示,包括:1)详细研究了压缩感知理论和稀疏表示理论,对使用稀疏表示分类进行了描述;对核心理论包括稀疏表示分类、测量矩阵设计和相对应的重构算法设计进行了深入的研究,并对其中一些算法进行了仿真和分析;2)对人脸识别理论的基础知识和算法进行了探索研究,结合一些新出现的理论方法,如压缩感知,对人脸识别理论中的特征提取和识别分类的一些方法进行了研究和仿真实验;3)将压缩感知尤其是理论中的稀疏表示,应用于人脸识别技术之中,并将传统的人脸识别算法和最新提出的稀疏表示分类算法做了一个对比仿真实验,通过识别率和识别时间得出结论,基于特征提取的SRC算法对遮挡、光照等外界因素的抗性较好,对外界干扰很鲁棒;4)针对SRC算法的识别速度慢这点不足,研究并设计一个结构化稀疏表示的人脸识别算法,即对图像进行分块处理、训练和识别,实验表明识别率有一定的提高,识别速度提高明显。结合本文设计人脸识别算法设计实现了一个基于MATLAB GUI的人脸识别系统,并结合了一个实际应用储物柜,设计了一个基于压缩感知人脸识别储物柜系统。更多还原

【Abstract】 The traditional identity recognition system is not enough safety with the development of science and technology,and the face recognition technology based on the biological characteristics is the hotspot and difficulty in building new identification system.Face recognition technology is a multidisciplinary cross theory.Compressed sensing is a new sampling theory,in recent years,based on the compressed sensing of face recognition technology is a new hotspot in the field of image recognition.In this paper,the sparse representation classification algorithm(SRC)for related research,and in view of some deficiencies of SRC algorithm made some improvements,builds a facial recognition system based on compressed sensing.This paper mainly do the following several aspects of research,including:1)Compressed sensing theory and sparse representation theory is studied in detail,the use of sparse representation classification has carried on the detailed description,The core of the three compressed sensing theory including the sparse representation,design and measurement matrix reconstruction algorithm design are studied,and some of the algorithm is simulated;2)Research on facial recognition theory,and argues that the traditional face recognition theory is complete,combined with some new theory and method,such as compressed sensing,study and simulate some method about the feature extraction and recognition classification;3)The compressed sensing used in facial recognition technology,and make a contrast simulation about the traditional face recognition algorithm and sparse representation classification algorithm is put forward by the latest,Through the recognition rate and recognition time come to the conclusion that the SRC algorithm based on feature extraction of shade,illumination has a good robustness;4)For SRC algorithm to identify slow this shortage,we design a structured sparse face recognition algorithm,namely to block processing,training and recognition of images,the experiments show that recognition rate is improved,the recognition speed increase significantly.Combined with facial recognition technology to design a face recognition system based on MATLAB GUI,and combined with a practical application storage cabinet system,designed a face recognition storage cabinet system based on compressed sensing.更多还原