Abstract: This article describes how to judge whether there are any faces in the video or image, if there are, it will count out the number of the faces. The principle of implementation is based on AdaBoost algorithm. This paper selectes Haar-like characteristics and trained cascaded classifiers to recognize the faces. The improvment is adjusting weight to every cascaded classifier dynamically, set heavy weight for cascaded classifiers with higher accuracy and low weight for cascaded classifiers with lower accuracy. Experimental results indicate that the method is fast and reliable and meets the requirement of realtime system.

Key words: AdaBoost algorithm, Haar-like characteristic, cascaded classifiers, face recognition