Research Of Face Detection Based On Skin Color Model And Adaboost Algorithm

Face detection is a process which uses certain search rules to determine whether itcontains face features for given input image or video and returns the face location and size ofthe face. Face detection technology has been widely applied in many fields such as facerecognition, video surveillance and human-computer interaction in recent years. It has becomea hot topic in the fields of Pattern Recognition, Computer Vision and Image Processing.Some classical algorithms on face detection are summarized and analyzed. Aiming at thedisadvantages of using a single method for face detection, this paper uses a combinationmethod based on skin color and AdaBoost algorithm. Using color information segment skincolor regions to obtain face candidate regions, then the improved weight update rule is used inAdaBoost algorithm to train cascade classifier. This method can achieve accurate positioningof the human face. The main work of this paper has the following three aspects:(1) According to the clustering characteristics of skin color information in YCbCr colorspace, skin color Gaussian model is established. It is used to calculate the skin color similarityimage and segment the skin area to get binary image by adaptive threshold method. Finally,we can use the method of mathematical morphology to process the binary image and utilizesome general face knowledge of human face to select these candidate face regions.(2) The source of AdaBoost algorithm and the related theories in the process of trainingare introduced. On the basis of Haar-like rectangle features prototype, we add the45-degreerotated rectangle features and introduce integration method to achieve rapid calculation forHaar-like eigenvalue. This paper analyses the issues of overfitting in AdaBoost trainingprocess and proposes a new weight update rule. This improvement can inhibit the excessivegrowth of the sample weight by setting right value update threshold. The method effectivelyavoids overfitting and improves the performance of AdaBoost algorithm in a certain extent.(3) Analyze the advantages and disadvantages of skin color face detection and AdaBoostalgorithm face detection, combine two kinds of algorithms to realize face detection. TheYCbCr color space skin color segmentation as the pretreatment of face detection reduce the improvement AdaBoost cascade classifier’s detection range. Experiments show that themethod is suitable for face detection in complex background with lower error detection rateand higher detection speed and robustness.