The Study On Small Sample Sets Face Recognition Based On Sample Augment Method

With the rapid development of computer application technologies, face recognition is paid more and more attention as a branch of biometrics. Over the years, many algorithms for face recognition have been proposed. Therefore, both the accuracy and efficiency of face recognition methods have got very great improvement. In a practical application, because of limited storage space and limited time to capture training samples, a face recognition system usually suffers from the problem of non-sufficient training samples. However, the problem non-sufficient training samples will influence the recognition rate seriously. In this study we will deal with the insufficiency of training sample size problem.Small sample sets problem refers to insufficiency of training samples lead to recognition rate declines or even cannot recognize. Different pictures of the same face might include various changes of expressions, poses and illumination. If the training samples are sufficiency, the facial feature s will be more fully included, and the recognition rate will be higher. If the training samples are non-sufficiency, the facial features not enough to contain all useful variations and the recognition rate will be lower. Generally, more training samples are helpful to achieve higher recognition ac curacy.For the problem of small sample sets face recognition, the study proposed a novel sample augment method to solve small sample sets problem. The proposed novel sample augment method fuses two parts of virtual samples to enlarge training samples which transforms the original face image to obtain “mirror” and “symmetry” virtual training samples; and then, calculate all scores which are generated by original training samples, two parts of virtual training samples and test samples; finally, combine all scores obtained in above step and obtain the final score. The experiments on various face databases show that the proposed method works better on small training samples size sets and outperforms over many respective representation-based methods in face recognition.For future optimize the algorithm, we use wavelet to reconstruct the training samples and the test samples so that keep useful features and filter irrelevant features. The improved algorithm performs better than the original algorithms.