| Face recognition is one of the most challenging topics in the field of pattern recognition and machine vision, and it has a great deal of applications in public security, information security, finance and so on. In the past several years, face recognition under the controlled condition has achieved great improvement. However, face recognition under uncontrolled environment is currently far from mature, especially pose variation is identified as one of the prominent unsolved problems in the research of face recognition. Therefore, face recognition across pose has important significance.The dissertation discusses the pose estimation in face recognition, proposing a method estimating the pose using Hidden Markov Model(HMM). By exploring the relationship between the face symmetry and pose variation, Fourier transform is taken as a feature extractor to extract the symmetry feature of the face. For further making use of the feature to estimate the pose, HMM with a strict and rich structure is used to model these features, obtaining pose models. Experimental results on the Weizmann databases and the CAS-PEAL databases show that the proposed method can improve the accuracy of pose estimation.The dissertation proposes a method of HMM and linear regression based face recognition across pose. On the base of the result of pose estimation, the dissertation discusses the problem of posture correction. By using the global linear regression(GLR) to learn the relationship between non-frontal faces and frontal faces, posture correction is transformed into posture prediction. To better recognize the face with pose variation, the GLR is extended to local linear regression(LLR). For the case of one training sample, the dissertation obtains additional virtual samples by projecting the single sample into different relationships. At last, HMM is used as a classifier to recognize the faces across pose. Experimental results show that the proposed method can achieve good recognition accuracy. |
PDF Full Text Request