Face Recognition In Natural Scenes And Its Application

Given the most important biometric feature of human — a face, the problem of recognizing who it is has received significant attention in the past decade. Identifying face images shot by various photographers at different time and different location, or face images targeted individuals at different occasions, is a very interesting but challenging topic. Thanks to its convenience and friendliness to users, face identification has been applied to many areas including access control, intelligent surveillance, medical record, friend recommendation and so on. Today, cellphones make taking pictures very easy and image amount surges in the internet. And those pictures taken by users are different from those taken under controlled environment. They vary from pose, illumination, occlusions and some other factors. So it brings in more challenges in face identification.We aim to find an algorithm to solve the problem mentioned above. Spatial-SIFT is used to represent the features of faces, and disparate local features demonstrate different discriminative power for face recognition. And then we learn a discriminative weight for each feature and feature couple. Those learned weights are then used to vote for its corresponding class. The proposed algorithm identifies the face of individuals with the spatial-SIFT feature voting under the image-to-class framework. Evaluation on two challenging Datasets shows that our method has achived the existing face recognition algorithms such as SRC and AHR.We use Visual Studio and Opencv liabrary to develope an automated face retrieval system on the Windows framework. The system consists of four modules: shot segmentation, face detection, face identification and face retrieval. The system can take a face image or a video sequence as an input and then it can return all the pictures and videos which include the input face image.