A Face Verification Algorithm Based on Negative Independent Sample Set and SVM

In many face verification applications, there are no face database for SVM training, such as PC face security logging-on system. A new face verification algorithm based on negative independent sample (NIS) set is presented, by analyzing existing SVM-based face verification algorithm and the demand for practical application. The approach generates enough positive samples by means of wobbling eyes of the user's registered image, employs FLD to extract feature, and deletes uniform samples in NIS with the rank-based FLD face recognition method. This scheme can resolves the classification conflict problem between the negative and the positive sample sets. After the negative and positive samples are sent to SVM for training, the SVM can do face verification for face image. The experiments on the SCUT face database indicate that the proposed method can ensure lower FAR if the negative samples are large enough, and the number of support vectors does not increase alone with the number of negative samples. A PC security logging-on system has been developed based on this face verification algorithm.