Research And Improvement Of Adaboost Algrithm In Face Detection And Recognition

Face detection and recognition are very active research directions in computer vision and pattern recognition fields, which can be widely used in some Safety departments, Intelligent entrance guard system and so on. Classical face recognition algorithms are good solutions. But when dealing with large-scale data, there are some shortcomings such as memory limitation, too long training time and other issues. After years of research, it appears many excellent algorithms which have good generalization ability. Adaboost is one of the excellent ensemble algorithms which can greatly slove the above problems. But when faced with a complex and large-scale data, how to improve the algorithm's performance, expand its applications are still problems to be studied.After making a deep research in the design of algorithm, classification criteria, classification performance of classical recognition algorithms and some variant Adaboost algorithm, this paper mainly complete the following work:First, we research some classical recognition algorithms including Principle Component Analysis (PCA), Linear Discriminant Analysis (LDA) and Independent Component Analysis (ICA). It is proved that these algorithms have good classification rate and speed. But the classification speed is not good enough and can also be improved.Second, we study the application of Adaboost algorithm in face detection which is proved to be a rapid and robust detection system. We chose simple haar features as the WeakLearner and do some experiments.The experiments show that this system has some shortcomings:error detection, undetection and overlap detection.we detailedly discuss the problems and give the solution.Last, We deeply research the application of Adaboost.Ml—the variant of Adaboost algorithms applying in multi-classes problems. The experiments show that Adaboost.M1 have good generalization ability, good classification accuracy and speed in face recognition. On the basis of such in-depth study, we propose a novel improved algorithm:change the ways of updateing weights in each iteration. It is proved that this novel algorithm have faster learning speed and with better classification accuracy.