The AdaBoost algorithm is not ideal in multi-view face detection,and the skin color model has a high false alarm rate under the complex background.This paper presents a method of rotating images form flank to frontal, which combines face detection based with the AdaBoost algorithm and the Skin Color Model.This method is mainly to rotate images, and makes different angle faces detected, then makes eyes detected, if the position satisfies the condition by calculating the position between the two eyes.A final verification is performed by the skin color model.The final key is inversion calculation,which draws the face location.Experiments prove that this method is adaptive to different views.%针对AdaBoost算法对多姿态人脸检测效果不理想和肤色模型对复杂背景下的图像误检率高的问题,本文将基于肤色的人脸检测与基于AdaBoost算法的人脸检测结合,提出一种由偏到正的检测方法.主要是通过旋转图片,使人脸分类器不会因为角度问题产生漏检,然后根据分类器检测出的两眼,计算两眼之间的位置关系,判断人脸是否处于正面位置,满足条件则经过肤色模型再次验证以后,对人脸位置进行反计算,计算出原图中的人脸区域.大量实验验证了本方法的有效性.
展开▼
