Research On Face Recognition Based On AdaBoost Algorithm

Face recognition is a biometric identification technology possessing great developable potential, as one of the hot topics in the research field of pattern recognition, image processing, machine vision, machine learning, and cognitive science in recent years. Face recognition has broad application prospects in areas such as security verification system, credit card verification, the public security system, identification of criminals, banking and customs control, human-computer interaction. Because face recognition is affected by many factors, the complexity of the problem itself, there are a number of key technology to be resolved, the study of face recognition is therefore of great theoretical significance and application value.This paper did some face recognition theoretical and experimental research focus on face feature extraction, feature selection and classifier design algorithm. The study in this paper mainly includes the following several respect:First of all, the current commonly used face recognition theories and methods are briefly summarized, the current status, problems and future development direction of the face recognition technology are then simply discussed.Secondly, in the aspect of feature extraction and feature selection, we proposed the Gabor Jet face feature selection method based on AdaBoost algorithm.Although face recognition based on Gabor features can withstand illumination and noise very well, having a higher recognition rate than other methods, but Gabor filter based feature extraction methods are normally computationally expensive due to high dimensional Gabor feature of face image, and let the algorithms not enough real-time. To address this issue, in this paper we proposed using the AdaBoost algorithm to select some of the key facial feature points, using the Gabor Jet feature of these selected characteristics face points to represent the face. Compared to Gabor face feature selection based on AdaBoost algorithm, the time of feature selection consumed could be greatly reduced; compared to Elastic graph matching algorithm, sub-optimal solution can be avoided by identifying the key feature points caused by the human factors. Finally, in the classifier design, proposed the multi-classification methods based on AdaBoost algorithms, Error-Correcting Output Code and support vector machine.Support vector machine used to be as a two-class classifier, including face recognition, most of the pattern recognition problems are multi-class classification, people proposed a variety of methods to solve a multi-class classification problem through assemble a number of two-class classifiers. In each iteration of the AdaBoost algorithm, the weights of the example that are wrongly classified are increased, so that in the next iteration the weak classifier learning algorithms are focused on the samples easily be wrong classified. Proposed multi-class classification algorithm based on AdaBoost Algorithm, Error-Correcting Output Code and support vector machine unlike the Error-Correcting Output Code and support vector machine in which all of the training example are involved in the training of the weak classifiers, but in each iteration, part of the sample are choosed to participate in the process of the weak classifier training according to the sample weights, so under the circumstances of a large number of samples, it can reduce the training time for each weak classifier, while making the training support vector machine classifiers more diversity, enhance the generalization ability of overall base classifier group.