CAJ | 학술논문

为解决当被检测图像中具有复杂背景或者含有多人脸时，不能够快速准确的进行人脸检测的问题，本文提出一种基于肤色分割和改进AdaBoost算法的人脸检测方法。首先利用肤色分割方法对样本图像实现图像的预处理，排除样本图像的复杂背景和人体非肤色区域，简化后续的人脸检测工作。然后对AdaBoost算法的弱分类器使用双阈值判决方法，以减少弱分类器个数，提升训练速度；改进权值更新规则，防止训练过程中出现过分配现象。最后对基于肤色分割得到的区域图像利用改进后的Adaboost算法进行最后的精确人脸检测。仿真试验表明，两种算法结合后在训练速度上提升，在检测速度和检测率上有明显提高。
위해결당피검측도상중구유복잡배경혹자함유다인검시，불능구쾌속준학적진행인검검측적문제，본문제출일충기우부색분할화개진AdaBoost산법적인검검측방법。수선이용부색분할방법대양본도상실현도상적예처리，배제양본도상적복잡배경화인체비부색구역，간화후속적인검검측공작。연후대AdaBoost산법적약분류기사용쌍역치판결방법，이감소약분류기개수，제승훈련속도；개진권치경신규칙，방지훈련과정중출현과분배현상。최후대기우부색분할득도적구역도상이용개진후적Adaboost산법진행최후적정학인검검측。방진시험표명，량충산법결합후재훈련속도상제승，재검측속도화검측솔상유명현제고。
To solve the problem that can not detect face quickly and accurately when the detected images has complex background or contain more faces,A face detection algorithm combined with skin color segmentation and improved Adaboost algorithm is presented.Firstly,pretreatment of sample images to achieve image by using skin color segmentation,exclusion of sample images with complex background and human non-skin regions,the follow-up work has been simplified.Then the dual-threshold decision method of weak classifiers Adaboost algorithm is used to reduce the number of weak classifiers and improve the training speed.Improved weight update rules to prevent distribution of the phenomenon appeared in the training process. Finally, the improved Adaboost algorithm is used to get the final accurate face detection based on areas of the image through skin color segmentation. Simulation results show that the combination of two algorithms improved the training speed, the detection speed and detection rate has improved significantly.