Multi-view face detection under complex scene based on combined SVMs

摘要

A single face classifier has difficulty in detecting multiview faces under real and complex scenes due to various poses, cluttering environment and small size of faces. In this paper, we propose a novel combination of SVMs to detect multi-view faces, using both cascading and bagging methods. In our method, the faces are divided into seven views. Each of them models a typical pose under complex scenes. By the modified bootstrap method applied in our method, a cascade of SVMs are constructed to quickly select face candidates from image with expected accuracy. Bagging of different SVMs can further eliminate the false detections that are difficult to handle by single SVM. Such combination of SVMs can effectively detect multi-view faces even with large rotation angles and heavy shadow. The experiment results show better accuracy and generalization performance over single classifier.