Research On Key Technology Of Face Detection And Recognition

Face image analysis and recognition is a research topic with high theory and application value.How to simulate the human visual characteristics,and use the computer technology to analyze the face image to obtain the desired information,attracting a large number of scholars to study from multiple fields such as image processing,machine vision,pattern recognition and artificial intelligence.This paper is aimed at several typical problems in face image analysis and recognition technology,such as face detection and location,feature extraction and classification recognition,to carry out research work,the main contents are as follows:Aiming at the problem of face detection in face recognition,a detection method which from coarse to fine combining AdaBoost classifier and skin Gaussian model in YCbCr color space is proposed.Firstly,the AdaBoost classifier based on Haar rectangle feature iterative training is studied,and it is applied to split candidate face regions in the image to be detected.A single Gaussian model of two-dimensional distribution of Cb-Cr chromaticity color component under YCbCr space is established.Based on this model,a skin pattern is built.Also the morphological filtering and the face priori knowledge are used to obtain the corresponding binary image,according to this binary image the face candidate regions is filtered to remove the non-face area.The simulation show that the method has good performance and low false detection rate.A feature extraction method based on two-layer two-dimensional discrete wavelet,principal component analysis and linear discriminant analysis is designed for feature extraction in face recognition.First of all,the detected face image is preprocessed to enhance the facial features of the face,then the two-layer two-dimensional wavelet is used to decompose the preprocessed image,and the second-order low-frequency image is selected.Also,the fusion PCA and LDA algorithm are used to extract features from this image.Finally,the Euclidean distance is used to achieve classification recognition.The simulation results show that the proposed method can effectively reduce the dimension of the data and fully retain the main feature information of the face,it also has some anti-noise ability.Aiming at the classification recognition problem in face recognition,a face recognition method based on improved genetic algorithm and BP network is proposed.The BP neural network is used to design the classifier,aiming at the shortcomings of BP network with low generalization ability,the genetic algorithm is used to optimize the weight of the initial network.To improve the selection operator of traditional genetic algorithm,optimal selection strategy is combined with the sort selection strategy,at the same time,the adaptive crossover and mutation methods are adopted to improve the merit ability and the self-regulation ability of the population,so that the convergence result is more likely to approximate the global optimal.The experimental results show that compared with the traditional face recognition method,the proposed algorithm has higher accuracy and robustness.