基于稀疏表示的人脸识别算法研究

发布时间：2018-07-25 19:47

【摘要】：近年来,由于生物特征具有不被遗忘与丢失的优势,所以生物特征识别技术便成为一种便捷、安全与可靠的保障技术。人脸识别技术应用计算机技术,利用人脸的生物特征来对人脸进行区分,被公认为最不具有“侵犯性”、最方便、非常“人性化”的技术。它已成为了现代计算机技术研究的热点之一。人脸识别主要包括人脸特征提取和分类识别两环节。本文首先系统地介绍了在人脸特征提取发展中产生的三个经典的算法:线性判别分析法(Linear Discriminate Dialysis,LDA)、主成分分析法(Principal Component Analysis,PCA)和局部保持投影算法(Locality Preserving Projections,LPP)。这三个算法都属于经典的降维算法,属于模式识别领域,极大地推进了人脸识别技术的发展,至今仍然被人们学习、效仿和比较。目前,基于稀疏表示的人脸识别方法得到了人们越来越多的重视,它从一个全新的角度看待和处理人脸识别问题。其基本思想可理解为:人脸测试样本可由训练样本近似表达,则测试样本属于表达式中占比最大的类别。本文应用稀疏表示的思想针对以上三种传统的经典的特征提取方法进行了改进,提出了三种基于稀疏表示的改进的常规变换方法(SRPCA、SRLDA和SRLPP),并对改进的常规变换方法做了合理性分析。人脸的分类识别已经有很多方法,近年来提出的基于稀疏表示的人脸识别方法属于一类新的人脸识别方法,它有着完全不同的方法学。本文首先介绍两个稀疏表示分类器:稀疏表示分类器(Sparse Representation based Classifier,SRC)、加权稀疏表示分类器(Weighted Sparse Representation based Classifier,WSRC),并在此基础上提出了经过改良的加权稀疏表示分类器(Improved Weighted Sparse Representation based Classifier,IWSRC)。最后,将本文提出的四个算法(SRPCA、SRLDA、SRLPP和IWSRC)在ORL、Yale、Yale B、AR等四个人脸数据库上进行实验,并对实验结果进行分析比较。实验表明,它们的性能比传统经典的特征提取方法和稀疏表示分别器都有了很大提升。
[Abstract]:In recent years, biometric recognition technology has become a convenient, safe and reliable technology because of its advantages of not being forgotten and losing. Face recognition technology applies computer technology to distinguish faces by using the biometric features of faces. It is recognized as the least invasive, most convenient, and very "humanized" technology. It has become one of the hotspots in the research of modern computer technology. Face recognition mainly includes two steps: face feature extraction and classification recognition. In this paper, we introduce three classical algorithms in the development of face feature extraction: linear discriminant analysis (Linear Discriminate), principal component analysis (Principal Component) and local preserving projection (Locality Preserving). These three algorithms belong to the classical dimensionality reduction algorithm and belong to the field of pattern recognition. They greatly promote the development of face recognition technology and are still studied imitated and compared by people up to now. At present, more and more attention has been paid to face recognition based on sparse representation, which looks at and deals with the problem of face recognition from a new point of view. The basic idea is that the face test sample can be expressed by the training sample approximately, then the test sample belongs to the category with the largest proportion in the expression. In this paper, the idea of sparse representation is used to improve the three classical feature extraction methods mentioned above. Three improved conventional transformation methods based on sparse representation (SRPCAS LDA and SRLPP),) are proposed and the rationality of the improved conventional transformation methods is analyzed. There are many methods for face classification and recognition. In recent years, the sparse representation based face recognition method belongs to a new class of face recognition methods, it has a completely different methodology. This paper first introduces two sparse representation classifiers: sparse representation classifier (Sparse Representation based Classifier and weighted sparse representation classifier (Weighted Sparse Representation based Classifier. On this basis, an improved weighted sparse representation classifier (Improved Weighted Sparse Representation based Classifier is proposed. Finally, the four algorithms proposed in this paper (SRPCAS SRLDAN SRLPP and IWSRC) are experimented on four human face databases, such as ORL Yaleer Yale Bearar, and the experimental results are analyzed and compared. Experiments show that their performance is much better than that of traditional feature extraction methods and sparse representation separators.
【学位授予单位】：苏州大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】