Study, The Fast Face Detection Algorithm Based On Adaboost

Face detection is a fundamental and important research theme in the topic of PatternRecognition and Computer Vision. As a key technology of facial information processing, ithas a broad application in many fields such as content-based image and video retrieval,video surveillance, automatic face recognition and human-computer interface, etc.Withlong efforts from the community of the face detection research, remarkable fruits havebeen achieved, among which statistics based methods hold a dominant position. Paul Violapresented a fast frontal face detection method based on integral image and AdaBoostlearning algorithm, which exhibits excellent performance in terms of detection speed andaccuracy.The thesis studies and analyses the face detection method presented by Paul Viola andgives some improvement of the original method. It improves the training process anddecreases the training time by stocking the categories of the face and nonface samples.Two critical factors in the method are discussed respectively: the manner of combiningeffectively the features and learning algorithm. Researches show that extension to thefeatures can largely improve the hit-rate and high-quality learning algorithm plays animportant role in decreasing the training time. Besides, the training dateset is alsodiscussed in this thesis. The experiment results show that a training dataset which containsfewer samples of side face can decrease the number of features in the cascaded classifier.Based on the inefficiency of "bootstrap", a new method is proposed, which can decreasethe time of selecting nonface samples largely by picking out a few nonface samples that aremost similar to the face samples.In the final, the thesis builds a rapid face detection system and compares it with othersimilar ones. It shows that the detection system in the thesis has a better hit-rate and lowerfalse-alarm-rate according to the detection results performed on the CMU Test Set.