The Research Of Face Detection Based On CMYK-H-CbCr Skin Color Model And Improved Adaboost Algorithm

Face detection is the basis of human face image analysis which includes face recognition,face tracking,facial expression and behavior recognition.In the past twenty years,face detection has become a hot topic in the field of computer vision.At the beginning of twenty-first Century,Viola proposed a face detection system based on Haar-like rectangle feature and AdaBoost algorithm,which was applied to real life for the first time in real sense.Viola face detection system,there are also some problems,first of all for high resolution images using the exhaustive search of face,reduce the detection speed of the system;secondly with the detection accuracy requirements increase,increase the number of classifiers,AdaBoost algorithm will be a fitting and "degenerate phenomenon" is very obvious.In order to speed up the detection speed and overcome the degradation.In this paper,a face detection system based on CMYK-H-CbCr skin color segmentation and improved AdaBoost algorithm is proposed.In order to balance the detection speed and detection accuracy,the system has been improved:A)Use skin color detection to filter most non face regions,and applies CMYK to skin color detection,and proposes a hybrid color space for skin color detection based on HSV and YCbCr.B)A new improved CS-LBP feature is proposed based on the improvement of CS-LBP features.At the same time,the improved CS-LBP feature is fused with the HOG feature,and a new method is proposed to describe the texture and contour of the human face.C)Skin detection after the candidate region will increase the AdaBoost algorithm "degradation phenomenon",based on the AdaBoost algorithm and the distance between the weight update process sample to sample weight threshold classifier to combine,which can improve the detection accuracy,but also speeds up the detection speed.Through the improvement,the proposed face detection system based on CMYK-H-CbCr color segmentation and improved AdaBoost algorithm respectively in MIT,Bio ID and AT&T database.The results showed that the system can improve the detection rate and improve the detection accuracy.