Face Recognition Algorithm With Reject Option

The face recognition technology has been very popular in computer visionresearch area, owing to its own advantage. With the increase in social demand andthe development of technology, more and more researchers devote themselves to theresearch region of face recognition. This research focuses on face recognition withreject option that means to refuse to recognize an object. In many real applications,such as an access-control system, the system should refuse to identify a person whois not registered in the system. To solve the reject option in face recognition, a facerecognition method with rejection function based on sparse representation isproposed in this dissertation.The proposed method classifies firstly the test set into two classes roughly: oneclass should be recognized normally and the other class needs be rejected. It cannotbe avoided that some samples are wrongly classified as the second class. In order toincrease the rejection accuracy of the proposed method, and find out those samplesthat need be rejected to recognize more exactly, LARK (Locally adaptive regressionkernel) and SIFT operator which is stable for the image rotation and scale transformare applied, are used to improve the accuracy of classification in the first step.LARK operator is robust to illumination variation of images. SIFT key pointsmanages well with pose variation of face images. To each sample that is classifiedas second class, we extract its LARK and SIFT features. Classify the sample basedon the two features, and get two classification results. At last, the sample isdetermined whether need be rejected to recognize by voting strategy.A lot of experiments on standard face databases, including ORL, YaleB and AR.At the same time, the proposed method is compared with other several facerecognition methods with reject option. The experimental results show that theproposed method guarantees a high recognition ratio as well as has higher rejectionaccuracy.