Font Size: a A A

Research On Face Recognition Algorithms Based On The Subspace Methods

Posted on:2010-04-16Degree:MasterType:Thesis
Country:ChinaCandidate:B H PengFull Text:PDF
GTID:2178360278459458Subject:Signal and Information Processing
Abstract/Summary:PDF Full Text Request
With the development of Electronic & Information technology all over the world, the fast and effective automatic face recognition (FR) system is concerned by the researchers. The automatic FR system is widely applied to National Public Safety, Social Security and business field etc. As we know, the face detection is one of the most indispensable parts of a FR system, and the accuracy of FR is affected by the face detection directly. However, the eyes as one of the most significant face features, whether we can locate the eyes accurately determine the rate and veracity of face detection. Usually, the dimension of face image is very high and the sample numbers are less than the dimension, which can result in Small Sample Size problem, and also bring large computation. Therefore, we use the subspace methods to reduce the dimension and for FR. In this paper, we researched the eyes localization methods and the algorithms based on subspace methods. Main content can be summarized as follows:1. In view of the inaccurate eye localization method, a rapid & exact eye localization algorithm is presented. The new algorithm combined the integral projection function with grads projection function. The effective and accuracy of new algorithm has been demonstrated by experiments.2. Considering the High-dimension and Small Sample Size problem, this paper research the traditional subspace face recognition methods. The experimental results show the performance of each algorithm.3. Researched the Discriminative Common Vectors (DCV) method. In order to reduce the computation, a DCV method based on wavelet decomposition of images is given, which firstly applied 2-D wavelet decomposition of face images and then used the DCV method. The experimental results show that the proposed DCV method based on wavelet decomposition achieve good performance and the computation burden is also reduced greatly.4. A new face recognition algorithm which combined the Principal Component Analysis (PCA) method with the difference common vectors is presented. Firstly, the common vectors are obtained by using Difference Subspace methods and consider them as the subtrahend vectors, and then the difference images of each sample are obtained. After that, the difference images can be viewed as new training samples and apply PCA face recognition method. The experimental results show the performance of the algorithm.
Keywords/Search Tags:eyes location, face detection, subspace methods, face recognition, Common Vectors (CV)
PDF Full Text Request
Related items