Research On The Method Of Face Detection Based On Skin Color Segmentation And Adaboost Algorithm

The definition of face detection is that detect the input image or video for faces, if itexists, then return the face position, size and posture. Face detection technology has importantapplications in many fields of video surveillance, intelligent security, video conference,human-computer interaction, the security access, etc. Its application range is wider than facerecognition. Therefore, face detection has become an important topic which attracts more andmore scholars.This thesis mainly analyzed the face detection methods based on Adaboost algorithm andskin color, and respectively improved two algorithms. After comparing of the advantages anddisadvantages of the skin color detection and Adaboost algorithm, a novel face detectionmethod combined the two methods is proposed. The research in this thesis main as follows:Firstly, this article studies the face detection algorithm based on skin color. Afterintroducing several commonly used color spaces and color models, we analyze the reasons forthe choice of YCbCr space and the Gauss model in this article; Give the process of skin colordetection, including image preprocessing, the skin color likelihood calculation, color thresholdsegmentation, morphological processing and color block selection; This article made someimprovement on the method of fixed threshold segmentation. Make the thresholdsegmentation by using Otsu algorithm which is more flexible and has wider range ofapplications than the fixed threshold segmentation. It can achieve a better effect.Secondly, the method of face detection based on Adaboost algorithm was studiedin-depth. This article firstly introduces the traditional algorithm, Haar-like feature, integralimage and the method of using the integral image calculate the eigenvalue, then analyzes thecomposition and training methods of the weak classifiers, strong classifier and cascadeclassifier, gives the flow chart of Adaboost detection process, finally, this article expands threerectangle feature that can describe more detailed and obtain more ideal effect than thetraditional algorithm.Finally, an improved algorithm combined skin color and Adaboost algorithm is proposedon the basis of the theoretical analysis of the experimental results and algorithm. Through a lotof experiments can prove that the improved detection method of this thesis can obtain higher detection rate and lower false detection rate than the two methods separately used, andimprove the detection performance.