Research On Face Verification Based On Multi-scale Local Feature And Global Feature

Face recognition includes many subjects such as image processing, pattern recognition, artificial intelligence, has been one of most active and challenging task in computer vision and pattern. Compared with other biometric technique such as fingerprint recognition and iris recognition, face recognition is more acceptable, get easier and cost-effective,which widely applied for security system, e-finance, daily entertainment. In this paper, we mainly concentrated on face verification and proposed a face verification algorithm based on the fusion of multi-scale local feature and global feature. Moreover, based on face recognition technology, a demo system with face detection, face alignment, face verification and face recognition functions has been designed and implemented.This paper has mainly completed the work listed below:1. Studied the common and classic facial feature extraction algorithm on LBP and HOG. We made a deep analysis and research on these two descriptors and pointed out that both of them are effective for face recognition. Therefore, we use LBP to extract local features because the excellent performent on texture measure. And use HOG to extract global features which is good at contour gradient description.2. Studied the subspace methods, such as PCA, LDA and SVM, analyzed their advantages and disadvantages and applicated in face recognition. Besides, we use Eigenfaces and Fisherfaces methods for face recognition in face database like Yale, ORL and GT, give the implementation steps.3. Proposed the algorithm based on the fusion of multi-scale local features and global features for face verification. We constructe the LBP locol features by extracting multi-scale patches centered at dense facial landmarks, which to enhance local features completeness and stability. At the same time, in order to reduce redundant information like hair, background, clothing, etc. we crop the face image to the appropriate size before extract HOG global features. Experimental results on LFW and GT face database show that the algorithm can effectively improve the accuracy and prove its feasibility and effectiveness.4. Implemented a face recognition demo system with OpenCV and Qt on Windows.The demo system has face detection, face alignment, face verification and face recognition functions, which user-friendly, easy to operate. Moreover, it can be easily applicated on other face database by replacing the train modules.