Comparison Research on Performance of Face Feature Extraction Algorithms Based on Sub-image Segmentation Abstract Face recognition technology is front-line task in pattern recognition domain, which can be used in all kinds of fields, such as security systems, criminal identifications, teleconferences and entertainment. Face recognition is also one of hot spots in pattern recognition and artificial intelligence. This thesis analyzes the actuality of automated face recognition. Combined with the previous research, two face feature extraction algorithms (coefficients of variances (CV) and singular value decomposition (SVM)) are proposed. Based on sub-image dividing idea and back-propagation neural networks, the performances of these algorithms are evaluated. In the face feature extraction algorithms based on CV, suitable coefficients are selected to denote images, which can reduce image information redundancy. The further research shows that the recognition rate can be improved and the computing recourse can be cut down if the sub-image including little information is removed. In the face feature extraction algorithms based on SVD, singular values of image have some good characteristics, such as stability, scale fixity and angle fixity. Compared with some algorithms, the proposed algorithms’ performances are higher. Based on ORL database, experiments results show that our algorithms have high recognition rate. Otherwise, a static face picture recognition system is developed based on the previous research. The system is realized by using VC++ 6.0 and theories of image proceeding and pattern reorganization. The system can recognize face image effectively and easily.
Key Words: face recognition, sub-image segmentation, coefficients of variances, singular value decomposition, back-propagation neural networks
