Rapid Face Detection Based On Learning

Human face detection is a process of finding the human face image extents and locations information of human face in a given image taking no account of 3D-pose, illumination and other conditions. Face detection is an arduous work mainly because extracting face features in face pattern is very difficult. Face features vary largely in different cases such as different human face poses , different illuminations , unknown lighting , occlusion and many other conditions.This thesis discussed the rapid face detection algorithm based on learning which is one of promising trends efface detection technica.There are three factors must research in the rapid face detection method based on learning: learning algorithm, the definition of features and the method of building classifier based on these features, the manner of combining effectively these features. This method firstly defines a type of feature format and then bases on it to build several weak classifiers using the method of building classifier whose performance is less good. Learning algorithm chooses and combines the weak classifier, generated the strong classifier combined by weak classifiers, and then it can be used to detect face. Most effectively combing the weak classifiers is important work for increasing the speed of detecting task. The thesis discussed respectively these three factors in the rapid detecting algorithm based on learning, studied emphatically the method of building classier based on features and the combing of the weak classifiers which is needed to build rapid face detection system. About building classifier based on Harr-like features, the thesis put forward a novel method based on the symmetry efface.In the final, the thesis build a rapid face detection system using a cascaded classifiers based on Adaboost learning algorithm, Harr-like feature and the method of building classifier. The final system detects image whose size is 320*240 in less than 0.1 second with detection rate of above 90%.