| Face Recognition Technology as one of the most successful applications of com-puter vision, has recently received significant attention and have reached a certain levelof maturity these years.The wide range of commercial applications and other nationalsecurity applications may account for this trend.Face Recognition Technology may be broadly classified into two groups which de-pends on if it make use of static image or of an videoclip.The Face Recognition Tech-nology which based on static image is also seperated into two subgroup,one is based onholistic approaches,just like eigenfaces and Fisherfaces.And another is based on Featuredescriptor,which relies on kinds of feature descriptor algorithms such as SIFT,one whoextracts a128-dimension vector of a keypoint to describe the feature,and BRISK,one whouses binary string to construct the descriptors.This thesis mainly focuses on three aspectand thus enhanced the accuracy of face recognition.1. First we introduced the SIFT algorithm and also its application in FRT.Followingwhich we come up with a novel face recognition algorithm based on SIFT.we car-ried out experiments on Extended Yale Faces B and also AT&T ORL databases andproved that our novel algorithm enhanced the accuracy of face recognition.2. Second,we introduced AdaBoost algorithm,and its application in human faces de-tection,following we proposed a new approch of FRT through expanding the orig-inal two-class AdaBoost to Multi-Class AdaBoost,and we carried out experimentsto support it.3. At last we introduced a new trend of feature descriptor:the binary-string-based fea-ture descriptor.we introduced two of them,BRISK and FREAK,after having anal-ysed their performances we come up with a novel algorithm we named it FGB-BK.The experiment carried out on EYB databases shows that FGBBK outperformsrecent state-of-the-art FRT algorithms while remaining simple and fast.At the end of this thesis we draw the conclusion and list future work we’ll make efort. |
PDF Full Text Request