A Study On Face Recognition Of The Nearest And Radial Basis Function Network

Face recognition is a challenging research subject, so that got many different disciplines researchers’s favor. Face recognition technology is widely applied to national public security, social security and business fields, such as public security case of criminal investigation, monitoring, information security, the company employees’attendance and so on.In face recognition system, it includes facial image preprocessing, face detection, feature extraction and face classification and so on. In the past decades, the researchers suggest many core of the algorithms of feature extraction and face classification, about these academic papers are endless.At present, the principal component analysis and independent vector analysis are the two main techniques of the face recognition feature extraction phase. To traditional PCA method, a PCA method based the singular value decomposition is presented, this method decreases storage space in feature extraction process. At the same time, the introduction of the energy parameter, through the choice of appropriate energy parameter,it makes the dimension of feature vector lower and can reconstruct the original high dimension face image, provides a good foundation for the classification of face image recognition phase.There are many methods of facial image classification, including based on geometric characteristics, based on template matching, based on the algebra feature, based on neural network, based on support vector machine and so on. Some face image classification methods combine with other classification methods, to meet the practical applications. Combining radial basis function neural network classifier and nearest neighbor classifier, this paper presents a fast nearest neighbor classifier.The radial basis function network is one of the neural network.It’s network structure is simple, training concise, learning convergence speed is quick and it could approximate any nonlinear function. This network is widely used in time series analysis, pattern recognition, nonlinear control and image processing. This paper uses clustering method based on supervision to determine the RBF neural network classifier structure.This classifier plays an important role in face recognition.It has a very high classification accuracy.The nearest neighbor classifier has high classification accuracy and good generalization performance.But the nearest classification algorithm exists a weakness, when the number of samples increases,the computation of classification also significantly increase, so classification speed decreases significantly. In order to overcome the shortages, this paper puts forward the fast nearest neighbor classifier. Through the research of the radial basis function network classifier principle, this paper make use of learning characteristics of the classifier to nearest neighbor classifier, for getting improved nearest neighbor classifier. Keepping in classification accuracy, that classifier increases classification speed. This method when classifying, through the condition judgment, step by step could rules out some training sample sets, only in the rest training sample sets finds target, which is nearest to testing sample, greatly reducing the nearest neighbor classifier calculation.