6396

5970

كارشناسي ارشد

نرم افزار

اصفهان: دانشگاه صنعتي اصفهان، دانشكده برق و كامپيوتر

1390

نه،130ص.: مصور،نمودار

ص.ع.به فارسي و انگليسي

محمدعلي منتظري

9/9/90

رسول امير فتاحي

مهندسي برق و كامپيوتر

ID5970

به فارسي و انگليسي: قابل رويت در نسخه ديجيتالي

Face Recognition by Fusion of Local Discriptors Mohammad Parsa m parsa@cc iut ac ir Date of Submission 2011 05 22 Department of Electrical and Computer Engineering Isfahan University of Technology Isfahan 84156 83111 Iran Degree M Sc Language Farsi Supervisor Mohammad Ali Montazeri Montazer@cc iut ac ir Abstract Today with widespread and increasing need for community safety in critical facilities and publicorganizations Methods for identification and verification of peoples become a key technology in theseplaces Such a need for reliable identification resulted in increasing attention to biometric technology Sincethe most common method for identifying individuals is their faces Face can be considered as one of the mostaccepted biometric criteria The face recognition can be considered one of the threads of research over thepast 20 years which attracted many researchers in various fields of science such as psychological science andcomputer science Many face recognition systems have reached the desired level of performance butdeveloping a face recognition system with good performance in uncontrolled conditions still remains ascientific challenge In such environments noise and lack of high resolution images caused a severe drop inperformance In this thesis we tried to develop a new face recognition system based on fusion of localdescriptors To this end two new local descriptors are introduced and thoroughly examined Then fusions ofthese descriptors in two forms are evaluated To reduce the high dimensional feature space resulting featurespaces are projected on LDA sub space corresponding to different areas of face image to produce adiscriminative face representation and robust to noise and low resolution face images Also for reducing theeffects of poor lighting condition on face recognition performance we have introduced a new pre processingmethod for illumination normalization of face images In this method poor illuminated face images areadjusted using dynamic range compression Afterward contrast of resulting image from previous step areimproved to have a good visually adjusted face image For evaluation of introduced face recognition system ORL and Yale datasets and for evaluation of illumination normalization method Yale B dataset are used Theresults obtained in experiments show that introduced face recognition system has ability to recognize facesunder severe noisy condition and low resolution images and can achieve good performance with few trainingimages Keywords Face Recognition Local Discriptors Linear Discriminant Sub space IlluminationNormalization

محمدعلي منتظري

رسول امير فتاحي