Fast Face Detection And Recognition Technology Research

With the rapid development of information technology, the technology of face detection and recognition is provided with extensive application value and theoretical value, which has become one of hot topic in the fields of image processing, artificial intelligence, pattern recognition and so on. Based on the study of related literature and the latest research results, technology theory of face detection and recognition is systematically studied in the thesis, which mainly includes two aspects of content: the skin color information based fast face detection algorithm and the linear space analysis based face recognition algorithm and its improved algorithm:1. About the aspects of the face detection, color information is used for rapid face detection. Firstly, an adaptive light compensation algorithm is used for the color image preprocessing; secondly, the polynomial model of standardized rgb color space is used for the rough detection of skin color area in the color images through light compensation; then, mixed skin model which is combined the polynomial model of standardized rgb color space with the gaussian model of the nonlinear transform color space of the YCg'Cr'is used for more accurate color regional extraction in the skin area of rough detection;finally, binary image the connected area is analyzed and elliptic area criterion is used to verify whether the face candidate area have face.2. About the aspects of the face recognition, the thesis studies the traditional principal component analysis and linear discriminative analysis, and mainly studies the face recognition algorithm of two-dimensional principal component analysis and two-dimensional linear discriminant analysis, then puts forward the improved algorithm and simulation. This thesis proposes a improved feature extraction algorithm which is simultaneously extracted form the rows and columns of the image, namely based on weighted (2D)~2-PCA face recognition algorithm and (2D)~2-LDA face recognition algorithm. The proposed algorithm eliminates the correlation of the rows and columns of the image, reduces feature numbers, reduces the storage space and increases the speed of recognition.3. About the face recognition, it often appears small sample problem because of the number of face samples is generally far less than the dimension of the sample of face image. So, a face recognition algorithm using Gabor wavelet transform is put forward in order to solve the small sample problem, namely each image of the sample image after the Gabor wavelet transform is regarded as independent face sample, which can greatly increase the number of every kind of face samples while the number of every kind face samples remains unchanged, and combined the improved LDA face recognition algorithm to form Gabor + (2D)~2-LDA face recognition algorithm.When the number of the training sample is small, the proposed algorithm can effectively improve the face recognition rate, and the performance of recognition is stable.