Muti-view Face Detection Based On Skin Color And AdaBoost Algorithm

Face detection has significantly advanced in the last decades. Especially in upright frontal face detection, current works have been developed that are able to achieve high detection rate and quick speed which has enabled their commercial applications. However, there are still many deficiencies in multi-view face detection due to the diversity of the multi-view faces. This paper investigates the multi-view face detection method based on AdaBoost Face detection method and skin color detection method. Two face detection methods are proposed based on skin color detection and AdaBoost algorithm,which can reduce the false alarms and improve the detection rate.Face detection based on AdaBoost algorithm which proposed by Viola has been widely recognized as the most efficient method. Based on this algorithm, I add new Haar features to train multi-view face samples. Skin color detection is easier to implement and much faster. This paper proposes two kinds of methods which combine skin color information with AdaBoost. Because these method can fast exclude non-face region, it can improve its detection rate and shorten detetction time.The main research work and contributions of this paper are as follws:(1) Based on the AdaBoost algorithm, add new Haar features to describe the multi-view face samples more accurate.(2) Select more than 10000 images form CAS-PEAL-R1 face database of Chinese Academy of Sciences and American FERET database. Then cut all the images to 20*20. These images compose a multi-view face sample database.(3) Using the AdaBoost algorithm as the core algorithm, train the multi-view face samples, combine the skin color detection to exclude much non-face region, finally achieve better result.In the end of this paper, I do lots of experiments to prove that the new algorithm can reduce the false alarms greatly while improve the detection rate or shorten the detection time.