Face recognition and retrieval based on feedback log information

Abstract

The robustness of the traditional face recognition methods is not strong, and the instantaneity and precision of the traditional face retrieval methods are not high. In order to dissolve these problems, a novel face image recognition and retrieval method based on feedback log information is presented. It first extracts the feature vectors by SIFT (Scale Invariant Feature Transform). These extracted features are invariant to illumination, rotation, translation, scale and expression. Then eliminates the false matching points according to the improved SIFT feature vectors matching criteria, calculates the distance between the face images with the residual matching points, and completes the first retrieve, with the result of returning the top 50 similar images. After this, refines the first result by relevance feedback strategy of the feedback log information to solve the semantic gap between the high-level semantics and low-level features of images. The experimental results show that, the proposed method has higher retrieval accuracy and more fast retrieval speed, and it is more suitable for real-time video investigation applications than the traditional relevance feedback.