Face recognition application on embedded system using support vector machines

Face recognition is one of the ordinary tasks of people; they perform it in their daily lives without difficulty. Human face has distinctive characteristic and human mind uses the visual information, which belongs to face, as biometric descriptor. Computer vision aims to identify people by using the face images; it imitates the complex function of the brain. Face recognition problem, uses a database, which contains face images of known people. When an input (face image or video that contains face images) is given to the problem; via this database the input can be defined and verified. FPGA (Field Programmable Gate Array) is frequently used in the fields that are digital signal processing, biometric identification, medical vision processing, space and defense systems, computer vision etc. FPGA is programmable logic units and users can be configured each logic block. In this thesis; FPGA and GPU based face recognition application is developed by using Support Vector Machines (SVM). First of all; Eigenface and Fisherface methods are used to extract features via MATLAB. This extracted features are educated by SVM, Artificial Neural Networks and k-Nearest Neighborhood methods and then their classification performances are compared. The system is educated and tested with ORL database that contains 40 people and each person have 10 different face images. During the education of the SVM 4 face images are used for each person and remaining 6 face images are used for testing. As a result, the success of SVM classifier has been reached 90% when we used eigenfaces and 91% when we used Fisherfaces. Keywords: Face Recognition, Eigenfaces, Fisherfaces, Support Vector Machines, Artificial Neural Network, k Nearest Neighborhood Algorithm