基于支持向量机的彩色图像人脸检测方法

提出了一种利用肤色信息、基于样本学习的彩色图像人脸检测方法。该方法利用两层支持向量机进行人脸检测,用肤色和非肤色样本训练的第一层支持向量机对图像中每个像素进行分类,所有被判断为皮肤点的像素构成了肤色区域;用窗口对肤色区域进行遍历,用人脸和非人脸样本训练的第二层支持向量机判断窗口是否包含人脸模式,并对检测到的人脸区域进行必要的合并。实验结果显示,本文方法对彩色图像中正面人脸的检测率为87.6％。

Face Detection in Color Images Based on Support Vector Machines

A face detection method in color images using two support vector machines combined in a hierarchical structure is proposed. The first support vector machine (SVM) trained with skin and non skin color samples classifies each pixel of the input image as skin or non skin pixel. Skin regions, which are the potential areas that faces may reside in, are composed of the skin color pixels. A search window is used to traverse through the skin regions. The second SVM trained with face and non face samples determines whether the search window contains a face pattern. Finally, a combination strategy to merge multiple detection results is performed. In the experiments, the method achieves a detection rate of 87.6% in color images containing upright faces.