Research On Face Detection And Recognition Algorithms Based On Static Images

As the rapidly growth of image processing technology and computing capability, Face Recognition Technology (FRT) is widely used in these areas, such as Intelligent Vision Surveillance, Entrance Guard and Checking-in System, etc. FRT is the most natural and direct identity verification measure based on biological characters, which is closely connected with computer vision and human-computer perceptive interaction.Face detection and recognition theories based on static images are studied in this paper, in which we research some algorithms with better precision, higher calculating speed, better adaptability and robustness. We improved face detection and recognition process relatively, in order to boost up the ability of face recognition system, followed by main research content:Firstly, aiming at the affecting elements such as illuminating intensity, photographing angles and exposing condition, which lead to disadvantages such as salt-and-pepper noise, variable illumination and lack of contrast in static images, median filtering method was used to preprocess the input static images. We also normalize training samples both from size and gray scale, save face character files and finish modeling work on ORL face database.Secondly, during face detection and orientation, we use Adaboost algorithm fused by skin color abstraction for facial characters extraction, and generalize statistical discriminating characters for differentiate individuals.Thirdly, during face recognition and classification, we analyse KPCA algorithm and improve LDA algorithm, then fuse KPCA and LDA, giving KFD (KPCA+LDA) method for recognition process. We compare detection and orientation results with face database, and list the most similar personnel according to the resemble values in matching step at last. Finally, the system uses OpenCV platform to make simulating and testing experiments for detection and recognition algorithms in VC++6.0 environment, and analyse experimental phenomenon and data. As a result, we know that the two combined algorithms researched in this article have both optimized face deteciton and recognition classifiers, making the system perform bettern in robustness under the condition of gesture rotation and facial expression, achieving relative balance in rate and velocity during face detection and recognition process.