Face Recognition Method Based On Gabor Features

As an important biometric identification technology, face recognitionhas become an active research in the field of biological recognition, andplayed increasingly important role in today’s automated identification.Face recognition emphasizes on face representation, the substance andhot spot issue of this research is feature extraction. Generally speaking,there are two kinds of face features, global and local. Global featuresare more sensitive to the impact of factors such as light, gesture, facialexpression. On the other hand, the localized feature reflects the localarea of the face, focus on extracting details of human face. So the localfacial feature can overcome the drawbacks of global features, and iswidely applied in face recognition. Gabor wavelet transform is a typicalmethod of local feature extraction, and has become a mainstream waydue to its superiority of facial feature extraction. The research of thisdissertation is conducted based on feature extraction by Gabor wavelettransform, and the main tasks include the following three aspects.(1) Firstly, the principle of two-dimensional Gabor wavelettransform has been analyzed systematically. In order to reduce the highdimension features of Gabor wavelet transform, the extracted featuresare down sampled. Then kernel principal component analysis are appliedto further dimension reduction. With regard to classifier design, supportvector machine is selected. On the basis above mentioned, an facerecognition method based on Gabor features is proposed.(2) To verify the effectiveness of the algorithm, the proposedalgorithm are compared with several current mainstream facerecognition algorithms on the ORL and Yale face database.(3) In order to further validate the practical application of theproposed algorithm, the theses has realized a face recognition systemusing MFC and OpenCV. This system includes of acquisition of the faceimage, image preprocessing, feature extraction and face recognition.By actual test of the face recognition system, we conclude that theproposed algorithm in this theses can be well applied in practical scenes.