The Research Of Face Recognition Method Based On HOG Feature And LBP Feature

With the progress and development of science and technology, the digital image has become an important form of multimedia. All aspects of our life involve a large number of digital images. How to identify and classify these images has gradually become a hot topic. In the face of huge amounts of digital images, manpower is not enough, so machine is needed to realize image recognition. Applying the knowledge of pattern recognition to the field of digital images can solve the problem. As an important member of image recognition technology, face recognition has broad application prospect, such as file management system, housing security credit card check, bank self-service ATM and so on. So research on face recognition is significant and has practical value.Firstly, we summarize the background, significance and research situation of face recognition technology. Then we give an account of basic knowledge of face recognition including the process of image recognition system, feature extraction and selection and pattern classification, which emphasize Eigenfaces, Fisherfaces, HOG feature and LBP feature. Finally, three classifiers, i.e. K-Nearest Neighbor classifier, Linear Regression Classifier and Sparse Representation Classifier, are described in detail. This thesis focuses on combining the excellent feature with the appropriate classifier to achieve high accuracy in face recognition. We explore the classification effect on the combination through the control variable method. The experimental result shows that:(1) Introducing HOG feature and LBP feature into linear regression classifier (LRC) can greatly improve the face recognition rate, especially in the case of less training samples, so it is with Nearest Neighbor classifier (NN) and Sparse Representation Classifier (SRC).(2)Although HOG feature and LBP feature are robust to variation such as illumination change, HOG feature has more advantages in response to obvious illumination change and expression change.(3) It is unanticipated to find the eigenfaces method not play a good effect on the image with obvious illumination classification.(4) Combining HOG feature or LBP feature with LRC can also improve the accuracy of face recognition, when we deal with occlusion in the face images.(5) After fixing each feature extraction method, we compare the performance of classifiers. We find that with the increase of training samples, to some degree, recognition rate with three classifiers is increasing. What is more, no matter what kind of feature we extract, SRC is the best classifier in the three classifiers on the whole. The second is LRC and the last is NN.