基于分段多高斯和辐射模板的人脸检测

Face Detection Based on Segmented Multi-Gaussian and Radial Template

针对现有的人脸检测方法中,平面内人脸检测准确率比较低和旋转角度误差比较大,提出了一种结合肤色、器官区域特征和辐射模板的新方法.首先,该方法采用改进的分段多高斯方法,分割出肤色区域.然后,在肤色区域内,通过区域相似性、肤色区域阈值化和人脸器官的几何性质,得到嘴巴和眼睛器官区域和这些区域的中心点.最后用辐射模板直方图,求得三个波峰和波峰之间的距离,定位人脸和人脸的旋转角度.实验结果表明,该方法不需要在肤色区域内移动辐射模板中心点和改变辐射模板的半径,能够在误差10°以内快速定位人脸.

In the current methods of face detection,the problem of the accuracy of face detection is relatively low and the error of rotation angle is relatively high in the plane, presenting a new method of combining the skin color regions ,organ regions and Radian Template Histogram. Firstly, this method adopts improved Segmented Multi-Gaussian to partition skin color regions. Secondly, obtaining the centers of mouth and eyes through combining the region similarity, the thresh-old of skin color regions and geometric properties of facial organ in the skin regions. Lastly, using the histogram of the Radian Template, Obtaining three peaks and getting the three distances between them, Locating faces and their rotating angle. The results show that this method can locate the face rapidly within the error of 10°, without moving the center of Radian Template or change the radius.