Research On Practical Face Recognition Algorithm

With the development of the field of pattern recognition, biometrictechnology such as fingerprint recognition, iris recognition, voice recognition,palm recognition and face recognition and so on has been applied to the fieldof authentication maturely. As a very important part of pattern recognition,Face recognition has become a hot research topic recent years, because of itsadvantage of no ease to disguise, convenience of recognition and good userexperience, which is applied to commercial field to meet the public demandof security. Generally speaking, automatic face recognition is accomplished inthree parts: face detection, facial feature selection and extraction, and facerecognition. Face detection is used to detect and locate the face appeared inthe image or video serials and normalize the face to the meet the standard ofalgorithm; Face feature selection and extraction is the method used to trainand select distinguishing features from the image and combine the featurespace used for further recognition; Face comparsion is used to verify oridentify a person from a database built formerly, of which most the is focus onstradegy of recognition in the aim of enhancing recognition rate.In this thesis, some key issues are primarily studied, aiming at praticaland real-time face recognition. And the main work of this thesis can bedescribed as follows:1) A fast eye localization methond based on a new Haar-like featureis proposeThis thesis presents a new haar-like feature generating the confidence of the feature throughout the candidate region in order to locate eyeballaccurately and rapidly. Our method is proved to be simple and robust againstthe disturbance caused by glasses, eyebrow and hair. The process of trainingand learning is not necessary because of the appropriate priori knowledge.Our experiment on three face databases shows that our method can be appliedto real time eye localization and even to pupil localization under mostcircumstances, achieving accurate results.2) Improvement of face recognition based on LBP patternThis thesis research face recognition based on LBP texture feature andproposed a face recognition method based on LBP with two layers. The facerecognition method, LBP with two layers, proposed in this thesis extracts theLBP features firstly, and then extracts the LTP features based on LBP featureswhich describe different location relation between LBP patterns. Finally,those two features combined, which refect different features of patterns areused for recognition. The experiments from standard face database show thatthe method presented by this paper can enhance the rate of face recognitionbased on LBP pattern.3) Improvement of face comparision based on multi-gallery andfeature selectionThis thesis creates a NIR face database in order to realize real-time facerecognition algorithm and researches on stradegy of face comparision basedon multi-gallery and LBP feature selection in order to enhance recognitionrate. In this thesis, two different methods are proposed compared with thenormal ways. Experiments on standard database show that our method candecrease the false accept rate while keep the accept rate.4) Implementation of face recognition system based on facerecognition algorithm Presently, the demand of face recognition is very large. In this thesis, aaccess control system based on LBP face recognition is realized in the aim ofoffice buildings and residential villas. Real-time experiment shows that thesystem has the property of real-time and accuracy.