Research And Implementation Of Real-time Face Recognition System Key Technologies

When doing face recognition, the extraction of facial feature is easily to beinterfered by external light. On the other hand, due to the changeable imaging angel andface expression, the performance of visible light-based face detection and recognitionalgorithm is not stable in different environment. Consequently, the key point of a stableface recognition system is to find the invariable facial feature in different lightenvironment. In this paper, we studied the general method of feature extraction andclassification in process of face recognition, and then proposed a new prototype of facerecognition system which is useful for daily guard of armed police. The main work andachievement follows:(1) To reduce the negative effect of inhomogeneous light on face which canseriously interfere the performance of feature extraction algorithm, weproposed a bilateral equalization method based on facial general structuralfeature to reduce the effect of deflection light, and gained the essentialcharacteristics of human face by detecting the facial edge information.(2) To benefit the time and frequency localization ability of wavelet transform, weused sym5wavelet basis to secondarily decompose image, and then calculatedeigenvalue by low frequency information of sub-band. This method is provedvalid by comparing with eigenface method in ORL and conducting experimentin variable light environment.(3) Based on VC++2008+OpenCV2.3.1+Access2003, the realization of a humanface detection and recognition prototype system, combined with armed policeon duty combat service needs, collecting samples of the new face recognitionexperiments, to obtain good experimental results.