Research Of Face Detection System Based On Skin Color Division And Adaboost Algorithm

Face detection is the important and first step of automatic face recognition system,how to quickly and accurately detect different faces in the target images directly affects the subsequent applications.In the field of computer vision and pattern recognition,face detection research has important theoretical significance,and in fields of the security,access control,human-computer interaction and so on also has extensive application value.For colorized images this paper will combine skin color feature and Adaboost algorithm to establish a coarse-to-fine detection system with two levels.In advance the first level of skin color division excludes a large number of non-face areas in complex background,determines the candidate region suspected of face,and then through the second level accurate detection algorithm based on Adaboost ultimately determines the faces,so as to improve the detection speed and system for face detection rate and reduce false detection rate.In addition,based on the previous research results of Adaboost,this paper optimizes some potential problems of classical Adaboost algorithm,so as to improve algorithm classification accuracy.The experimental results show that the face detection system in this paper has good detection performance.The main work of this paper includes the following aspects:Firstly,systematic comparative analysis of the basic theory in recent years and the mainstream methods of face detection are deeply studied and discussed.Secondly,this paper discusses the principle of face detection of skin color in the YC_bC_r color space and establishes a single Gauss skin color model,and uses threshold segmentation and morphological operations to get parts of the complete clean skin as the candidate face region.The adverse effects of light on skin detection are discussed and some solutions are given.Moreover,a detailed analysis of the face detection system based on Adaboost algorithm in all aspects of technology is discussed:algorithm principle,feature calculation,training process,cascade structure and the detection mechanism.Then this paper discusses Haar feature types and their extended templates,in addition the numbers of features are reduced,then focusing on the analysis of the Adaboost algorithm performance,this paper gives the multiple weight balancing strategy to solve problems of training sample weight in the unbalanced distribution;for weak classifiers which are similar,the method of difference maximization of weak classifiers is put forward.Finally,the advantages and disadvantages of the two methods of face detection based on skin color division and Adaboost algorithm are analyzed,and the two level face detection system based on skin color division and optimized Adaboost is designed.Experimental results show that this method can improve the detection speed,detection rate and reduce the false detection rate,and it has a positive significance to improve the performance of the detection system.