| Automating face recognition system is a popular research task thatdeals with many fields such as pattern recognition ,digital imageprocessing ,computer vision and neural networks .It can be widely appliedto security department ,television and telephone conference ,individualidentification ,digital control .The list goes on and on .Because of therapid increasing applications and recent development in the technologiesof neural networks ,wavelet analysis ,computer graphics and computervision ,it has received significant attention and active research .Manyapproaches about two-dimension gray image have been proposed ,which canbe divided into geometrical feature based ,eigenfaces based ,localfeature based ,skin-color feature based ,elastic matching based ,templatematching based ,neural networks based and hybrid approaches have beensuccessfully applied in face recognition system ,but they have advantagesand disadvantages.In this thesis ,the main works is faces recognition and it includetwo major part:extracting algebraic feature from a grey imagespost-processed and dynamic bi-threshold cumulation average sequencerecognitionThe extracted algebraic feature is based on Turk and Pentland'seigenfaces for recognition in this paper ,but there are many defect aboutthe method of Turk and Pentland : ①It relies not only on relation betweentraining image and tested image in essential ,but also requires toresemblance between training image and tested image. ②Although Turk andPentland's algebraic feature has been undergone projection transform anddecrease dimension ,there are feature vector on every image ,so everyindividual have several feature vectors ,hence influence recognitionspeed. ③because there are only several feature vectors everyone ,it cannot contain all feature and decline recognition rate.In this thesis , The method of Turk and Pentland's feature extractedis modified . The feature of each facial image is averaged by using the... |
PDF Full Text Request