Face Recognition Method Based On DE

With the development of information technology, face recognition is widely used in information security, security systems, human-computer interaction and other fields. Face recognition is a multi-disciplinary technology, involving a number of areas, which include image processing, computer vision, pattern recognition and so on. Because of the broad development prospects of the face recognition, face recognition is taken seriously by domestic research staff and business. And face recognition has great value of research.Face recognition has developed over several decades, and most face recognition systems have reached the recognition rate as high as 90%. But in practice, face recognition faces enormous challenges, including illumination change, attitude change, ageing change and so on. In order to achieve robust face recognition, feature extraction and classification can be studied. Extract the effective features of face to get the stable features in a changing environment. Wonderful classifier design will be more robust to change of face. And we can try to solve the problem in two ways simultaneously.This thesis mainly studied feature extraction and classifier design methods for face recognition. Major work and results are listed as follows:Several representative feature extraction methods of face recognition were discussed. Local binary pattern (LBP) and center-symmetric local binary pattern (CS-LBP), an improved version of LBP, are robust to facial expression and other changes. Gabor transform can extract multi-scale and multi-dimensional spatial frequency characteristics of specific area of the image, and it is robust to the changes of face. Combining the Gabor transform with LBP and CS-LBP descriptors, two mixed descriptors, the local Gabor binary pattern (LGBP) and center-symmetric local Gabor binary pattern (CS-LGBP), were obtained. The characterization of these descriptors were analyzed and compared through experiments.The classical multi-classifier ensemble learning algorithm, AdaBoost, was implemented, its effectiveness and efficiency are testified via experiments.A multi-classifier ensemble algorithm based on differential evolution (DE) (DE-MCE) was studied. The algorithm used differential evolution algorithm to optimize the weights of multiple classifiers to obtain an accurate strong classifier in relatively less time. The face recognition experiment results showed that DE-MCE algorithm achieved as high recognition rate as AdaBoost algorithm in less training time.