Face Recognition Based On Gabor Wavelet Transform And Support Vector Machine

Automatic face recognition is an identification technology that it analyzes face image with computer for getting effective recognition information. It can be applied in many aspects, such as criminal recognition of police system, driver's license, passport, monitor system of bank and custom, automatic doorman system and family entertainment etc.This paper mainly studies in two parts of face recognition: feature extraction and pattern classification.For feature extraction, this paper first implements three typical methods: PCA, Fisherface, Kernel Fisherface, analyzes and experiments with them.Then mainly probe into the feature extraction method of Gabor wavelet transform. Its advantage is that Gabor wavelet function can describe the receptive field of simple cells in the visual cortex of human cerebra correctly, so it can extract the face feature effectively.Finally this paper improve the feature extraction method based on Gabor wavelet transform. First getting Gabor wavelet transform feature by Gabor wavelet transform to face image; because the dimensions of transform feature are very high, so they must be reduced. This paper uses two steps to reduce the dimensions of Gabor feature vector, that is first to sampling the transform feature, then to reduce the dimensions again with an improved Fisherface that this paper gives for getting transform feature base. Projecting the face image onto the transform feature base, transform coefficient can be used to be the input of classifier.About classifier, this paper mainly applies a multi-class classifier composed of two class SVM classifier with error correcting. This classifier has error correcting ability because it uses the error control encode in communicational channel. Because of its error correcting ability, though several two-class classifiers have error output, they can not affect the final classified effect.Therefore this classifier has better classified effect than traditional classifiers. Simulation experiment with Gabor wavelet and the classifier gets 97.8% recognition rate in ORL standard face database...