The authors develop and test a multiclassifier-based near-real-time face detection system based on the premise that a three-part strategy is necessary for designing real-time face detection systems that provide high detection rates. The critical factors for real-time face detection are based on a framework of multiple classification functions: (1) a skin classification function is used as the preliminary stage in order to prune the search, localize the computation, and therefore improve performance time; (2) subsequently, three coarse-to-fine statistical model based classifiers are used to scan the windows and discard most non-face windows; and (3) finally, faces and non-faces are verified from images. The bagging ensemble algorithm (bootstrap aggregating) is also applied to improve the performance of detection rates.
展开▼
