Face Recognition Analysis Based On Two Dimensional Image Representation

Face recognition is a popular topic inthe field of biometric identification,and research on the face recognition technology has great theoretical and practical significance. Research into face recognition has flourished in recent years due to the increasing need for robust surveillance, security and rapid growth of Internet applications, and so on. But its recognition precision in practical applications still cannot satisfy the expected demands of people, especially under the condition of variations of ages, illumination and facial expressions, photographing azimuth or other disturbance existence in image. This thesis is in the framework of face recognition and focused onthe study on two dimensional linear feature subspace and imagetrans form approaches, such as two dimension principal component analysis(2DPCA), two dimension linear discrimination analysis(2DLDA), two dimensional locality preserving projections(2DLPP) and two dimensional nonlinear feature subspace and image transform approaches, such as the two dimensional discrete cosine transform(2DDCT) and nonlinear feature subspace and image transform approach, such as compressed sensing(CS). The research work of this thesis focuses on several key technologies including image preprocessing, facial feature extraction and face recognition algorithms. Some specific problems are discussed and some improved algorithms are presented in this thesis. Study on this field still goes on. For a thorough survey of the face recognition technology, this thesis first introduces the content, research significance and general progress in China and abroad. Then, the current typical face recognition methods are brief summarized. Finally, research on the current status of face recognition technology, problems and technological development are discussed.The main creative works in the dissertation are as follows:1) We proposed Double Sides2DPCA algorithm via investigating the two dimensional principal component analysis(2DPCA) and the generalized low rank approximation of matrices(GLRAM) algorithm. Experiments showed that the Double Sides2DPCA's performance is as good as2DPCA's and GLRAM's. Furthermore, the computation cost of recognition is less than2DPCA and the computation speed on eigen value is faster than that for GLRAM. Further, we presented a new constructive method for incrementally adding observation to the existing eigen-space model for2DPCA, called incremental2DPCA. An explicit formula for such incremental learning is derived. Experiments illustrate the effectiveness of the proposed approach. 2) Two Dimensional Linear Discrimination Analysis (2DLDA) is an effective feature extraction approach for face recognition, which manipulates on the two dimensional image matrices directly. However, some between-class distances in the projected space are too small and this may produce a large erroneous classification rate.We propose a new2DLDA-based approach that can overcome the drawback in the original2DLDA. The proposed approach redefines the between-class scatter matrix by putting a weighting function based on the between-class distances, and this will balance the between-class distances in the projected space iteratively. In order to design an effective weighting function, the between-class distances are calculated andthen used to iteratively change the between-class scatter matrix. Experimental results show that the proposed approachcan improve the recognition rates on benchmark data bases such as the ORL database, the Yale database, the YaleB database and the Feret database in comparison with other2DLDA variants.3) The two dimensional discrete cosine transform(2DDCT) features in frequency domain are more robust to the variations in illumination and rotation than gray-level data in time domain, and the2DDCT transform can retain the essential content of the information. The Two Dimensional Locality Preserving Projections(2DLPP) can remain the facial manifold structure in a subspace.Combination of the two features, the algorithm of2DLPP in thefrequency domain was proposed.4) The compressed sensing (CS) needs to pre-expand the face image matrix into a one-dimensional vector in the image recognition process, and the converted one-dimensional vector dimension is generally higher, so, higher resolution face images must be pre-cropping before using CS for face recognition.2DDCT transform can filter the nonsensitive MF and HF part of the image, and retain the essential content of information, so, we proposed CS in frequency domain for face recognition, the algorithm can achieve ideal identification rate in face recognition.