基于差值图像及肤色特征的人脸定位方法

该文提出一种用于安保监控系统中提取人脸的实用有效方法。该方法利用运动序列图像之间的差分图像，检测出运动目标区域，得到运动目标子图像，剔除大部分静止背景的干扰。随后对于图像进行肤色映射，得到二值图像，对二值图像进行水平和垂直投影，确定人脸的左右边界和上下边界，从而得到人脸的准确定位。该方法在工程上有效且计算量较少，可以达到实时需要。

Human Face Localization Based on Image Difference and Skin Color

We propose an effective method of locating human face, which is applicable to security systems. Motion regions in a motion image sequence are detected based on image subtraction, and sub images containing head and shoulders are obtained. Disturbing background is then eliminated. The sub image is then mapped to a binary image according to the skin color map. After a series of morphological operations, the binary image is projected to the right, left, top and bottom. In this way, the head and face are located. The method is practical and can meet the real time requirement.