| Automatic face recognition (AFR) aims at endowing computers with the ability to identify different humans according to face images, which is a hot study topic in the field of pattern recognition. A number of AFR systerms have been applied to security authentication and human-computer interaction. However, when it comes to the variant illumination, pose and expression conditions, the performance of AFR systems need to be improved. The main results obtained in this thesis are as follows:(1) In order to improve the detection rate and simplify the AdaBoost detector structure, an algorithm named Real AdaBoost based on LAB is proposed. A YUV histogram model is presented to reduce error detection rate caused by gray scale image.(2) Two subspace-based methods that can extract discriminative information from low dimensional space are designed and implemented with PCA and LDA. However, experimental results show that subspace-based methods are not robust in single-sample situation.(3) In order to solve the referred problem of subspace-based methods, a face recognition method based on Gabor filter and local binary pattern (LBP) is introduced. The approach extracts LGBP histogram sequence from single picture to obtain robust face representation. Experimental results show that the novel method is robust and can alleviate the impact caused by shortage of samples.(4) After analyzing the characteristics of automatic face recognition systems, an automatic face recognition system based on modified AdaBoost and LGBP is built.A number of experiments have been carried out to evaluate the performance of the automatic face recognition system, the results of which demonstrate that face detection based on real Adaboost and LAB can get better performance than before, and error rate can be reduced by verification with YUV-Color model. Face recognition based on LGBP histogram sequence can achieve higher correct rate with fewer training samples. |
PDF Full Text Request