采用图像差分法先检测出运动区域,消除了背景对人脸检测的影响,缩小了搜索的范围。利用肤色在YCrCb空间的聚类特性来检测人脸,同时为了去除亮度对肤色的影响,增强检测的可靠性,采用了亮度补偿。最后选取眼部特征作为条件确定人脸区域。实验表明,算法能准确地检测出人脸,并具有一定的实时性。
This paper proposed a real-time method to detect faces in video. At first we use image difference method to detect motion area, which eliminates the impacts on face imposed by the background and narrows search area. The clustering character of cnmplexion in YCbCr space is applied to detect face, and at the same time ,light compensation is used for wiping off the effect of light on complexion and enhancing the reliability of the detection. At last the eye features are chosen as the determination condition for true faces. Experiments show that this algorithm can accurately detect faces from image, and its real-time property is pretty good.