A Cascaded Face Detection Framework

J. Zhu, S.C. Schwartz, and B. Liu (USA)


face detection, computer vision, pattern recognition


Face detection is an important basic technique for a variety of image processing tasks. To improve its applicability, we study fast face detection techniques employing low complexity algorithms. This paper reports on a cascaded face detection framework that incorporates a feature-point detection component, and a probabilistic modeling component based on a subregion division scheme. Finally, a verification component that is employed to improve the performance of the whole system is discussed. Feature point detection takes advantage of the ad hoc knowledge of the features in an upright frontal face. By detecting those points frequently appearing in the area of eyes and mouth, and then heuristically and sequentially screening out many input regions in an image based on the geometric features, this step speeds up considerably the detection process. The low complexity statistical modeling of the subregions in a face, i.e., three horizontal strips and nine rectangular blocks, further excludes some candidate face regions. Experimental results demonstrate that this low complexity cascaded system has good detection performance.

Important Links:

Go Back