Research On Face Detection By AdaBoost Algorithm Based On Joint Integral Histogram And New Haar-Like Features

With the rapid development of computer vision,the face detection technology has been widely applied to various aspects,such as eletronic products,security monitor and verification of identity.Frankly speaking,we often take advantage of facial features to unlock,to login,to pay,and even to face Social Security Authentication.Face detection is the primary technologie.It has made certain achievements.However,when facing complex or unfavorable testing environment and a variety of scenarios with increasingly high performance,the existing algorithms still can not meet the demand.Hence this paper proposed AdaBoost algorithm based on Haar features,characteristics of the Multi-Block Local Binary Pattern and Joint Integral Histogram(JIH),which is called AdaBoost algorithm based on MB-LBP features,AdaBoost algorithm based on MB-LBP features,new Haar templates and JIH.It has a higher detection rate and a lower false detdction rate especially in those pictures where someone has side faces,or looks down,or wears sunglasses and where there is a strong interference background.Firstly,the artcle describes the applications significance and research status of face detection technology at home and abroad.It includes the introduction of the existing mainstream face detection algorithm and analyzes the advantages and disadvantages of the classical algorithm's principles.This paper focuses on the AdaBoost algorithm,Haar features and the process of classifier training.It extends the existing Haar feature set and the proposed new Haar feature set improves the detection rate of the system.However it still has some limitations.Secondly,the paper introduces the MB-LBP feature in the human face detetion which are invariant to rotation and robust to illumination change.It can utilize information of both whole image and local sections and eliminate the random noise.Therefore we propose a new method which combines MB-LBP features and Haar features to solve the issue in those situations where Haar feature are sensitive to illum ination variation Thetheoretiad analysisand experimentresulthave proved this method can not only solve the above problem,but also can further reduce the false rate.However the detection rate become lower and the testing time become longer,especially in the pictures which have a strong interference baceground.Finally,we use JIH function to combine MB-LBP funtion with the integral image function.We use new Haar features to extract new feature and then design a brief MFC interface to display results.The analysis and experiment show that the new JIH method can effectively avoid false detection in those pictures where someone has side faces,or looks down,or wears sunglasses and where there is a strong interference background.