Face recognition with Gabor phase

Abstract

Face recognition is an attractive biometric measure due to its capacity to recognize individuals without their cooperation. This thesis proposes a method to dynamically recognize a facial image with the help of its valid features. To validate a set of feature points, the skin portion of the facial image is identified by processing each pixel value. Gabor phase samples are examined, depending on whether they are positive or negative at each filter output, and feature vectors are formed with positive or negative ones along with the spatial coordinates at the validated feature points. The collection of feature vectors is referred to as the feature vector set. The face recognition system has two phases: training and recognition. During the training phase, all images from the database are automatically loaded into the system, and their feature vector set is determined. When the test image arrives at the system, the feature vector set of the test image is compared with that of database images. Feature vectors are location-specific, and thereby similarities between the feature vectors of the test image and database images are calculated, provided that they are from the same spatial coordinates. Once spatial coordinates are matched by using exclusive-OR (X-OR) operation, the similarity is calculated from the values of the feature vector. Simulations using the proposed scheme have shown that precise recognition can be achieved.