Design And Application Of Face Detection And Recognition System

This paper sums up the main methods of the face recognition in recent years. Besides, by combining the Gabor wavelet filter technique, wavelet analysis, generalized principal component analysis (generalized PCA) and radial basis function(RBF) probabilistic neural network, we construct a face recognition classifier and have a test using the ORL facing images database.Finally, we get a relatively high recognition accurate.The recognition classifier structure and operational procedures achieved in this paper are as follows:Firstly, choose from the center frequency parameters in the five directions and use all parameter information of8directions to construct the wavelet filter after which filters the facing images to extract invariant feature information of different frequencies and directions; Secondly, use wavelet decomposition to reduce dimensions on the Gabor feature facing images. Compared with the sampling method, wavelet analysis can effectively cut down the lose of feature information owing to its characteristic multi-frequency analysis.Then,do further dimensional reduction and extract the data feature by Generalized Principal Component Analysis method. Regard the figures which are gained by projecting the Gabor feature facing images into the featured subspace as the feature data and combine them into a column vector to be the input of neural network. Finally, take advantage of the outstanding clustering and classification abilities of radial basis probabilistic neural network to achieve facing classification and recognition.In this paper, OpenCV and VC++2008are used for face detection and recognition. The system is constructed by four step:The first step is image processing, for the picture may be too bright or too dark, it is first processed with color balance and light compensatio.The second step serving for skin region detection, the processed image is transformed into YCbCr color space through the clustering features of skin. Then, skin area is selected and the rest pixels will not be detected in the following step to reduce the computing pressur. Further more, dilation and erosion are used to reduce the burrs and noises in the images.The test results shows the algorithm performs well on ORL face data set. The system is tested by the practical pictures, the results show the system are roubust on detecting and recognizing frontal faces.The final step is the recognization step, which use Gabor wavelet algorithm with ORL data sets. The test results shows the algorithm performs well on ORL face data set. After examined by the practical pictures, the system has shown satisfying results while detecting and recognizing frontal faces.