Research Of Intelligent Human Face Recognition Technology Base On Hidden Markov Model

Face detection and recognition technology is one of the most significant and practiacal topics in fields of pattern recognition and computer vision. It aims at giving computer the ability to recognize and identify faces in image just like human being can do. On the base of a deep comparation and analysis of current popular face detection and recognition technologies, this dissertation discussed how to improve the accuracy and speed of the facial detection and recognition system, at the same time guarante that the system has a strong robustness, finally the author realizes a facial detection and recognition system.Through many testing experiments and a statistical analysis of the experiments's results, it is proved that the facial detection and recognition method mentioned in this thesis can achieve the goals which were listed above.This dissertation focused on the follow contents:In the discussion of facial detectioin: Combine face skin-color model with auto- adapted threshold selection method, use this threshold realize accurate segmentation of the skin color areas, after that , the follow pocesses were carried on in order to improve the image's quality: first using average value filter to detele the image's noise, then using some morphological image processing methods such as erode , dilate to cut the bridge areas which connectes different skin-color regions.eliminating the skin-color connecting areas which are not according with the human face proportional regulations, finally using template-match method to carry on the appraisal of the candidate face's match probability, some of them can be confirmed as human face if their probability pass the threshold. The experiment's result shows that: this method could select the optimal threshold considering the image's skin-color and background area's histograms. Which effectively enhance the accuracy of face detection by the accurate segmentation of the skin color area.In the discussion of facial recognition: transform the original image into frequen cy field through DCT. Extract face feature in frequency field, use every dct matrix's low frequency section as observer vector to train P2D-HMM, which can modeling human face very exactly; experiment result shows that it could achieve a very high recognition rate using this method.