Face Similarity Between Two Pictures

With the development of web2.0, more and more people like to share their favorite pictures, the Internet has emerged a broad array of image information, and how to obtain the desired information from Internet quickly and accurately becomes critical. As computer vision and image processing technology getting mature, getting the picture quickly and accurately becomes a reality, more and more Internet companies have also introduced the image processing applications. As for image recognition and classification problems, face recognition and categorization is one of the most valuable applications.This paper introduces the main algorithms and techniques of face recognition. According to face similarity between two pictures, the paper introduces the system design. The steps of face recognition are:Preprocess on the picture to get a face quotient image, use face detection technology to extract key points of face, face correction based on the key points on human face, at the same time cut12grids. Extract LBP, LQP and SIFT featrures, using the auxiliary data sets to get low-level feature. Based on low-level feature, train the feature using SVM classifier, and the predict results are recognized as high-level features. Then the high-level features contribute to the final classification.The precision of the algorithm is excellent. Using the most mainstream face recognition data sets--LFW’s View2set, true positive rate/false positive rate are0.83/0.1, which rank top ten on LFW’s official web site, proving the robustness and feasibility of the system. Dealing with download images on-line, the system also maintains a high accuracy.The next step is to analyze the theory of face recognition factors from the image processing and machine learning perspective. Reduce the impact of these factors and improve the accuracy, at the same time research on the low-dimensional features, reduce time consuming to meet face recognition requirements.