Based On Adaboost Algorithm Automatic Face Detection And Recognition

The AdaBoost (Adaptive Boosting) algorithm, introduced in 1995 by Freund and Schapire, is an algorithm to detect faces very quickly, which is a landmark advance in the field of face detection. This algorithm adjusts adaptively the errors of the weak hypotheses by the feedback of weaklearn, which advances detection accuracy rates greatly and doesn't reduce detection efficiency.This research, on the basis of studying the AdaBoost algorithm thoroughly, has collected lots of human face samples and non-face samples and trained a cascade classifier of 16 strong classifiers which has been made of by 1152 weak classifiers.In order to conquer the defect of lower detection speed when the AdaBoost algorithm has tremendous computations, the research has embedded improvements in front of classifier in the process of face detection. The improvements has tow parts: firstly, before the image inputted the AdaBoost algorithm, applying the skin color model RGB which needs fewer computations to detect the skin color pixels and finding the maximum face area in the image; secondly, in front of the AdaBoost algorithm using skin color integral image to abandon the sub-windows which having lower proportion of skin color pixels, which decrease the burden of classifier greatly.Because some information exist between tow frames in dynamic sequence images, but the face detection algorithm for static image has not used the information, we have used face track to forecast the position and area of the face in the next frame. This method has advanced the speed of face detection, and satisfied the timely demand of dealing with dynamic sequence images.Lastly, the research has used the method based on the geometry feature to identify the face detected.