基于肤色分割与AdaBoost算法的人脸检测

Face detection based on skin color segmentation and AdaBoost algorithm

提出了一种将肤色分割与AdaBoost算法相结合的人脸检测方法。该方法利用肤色的聚类特性在YCbCr色彩空间中建立肤色高斯模型,通过形态学处理完成图像肤色区域的筛选;然后利用AdaBoost算法训练弱分类器并构成强分类器,对强分类器进行组合,形成级联分类器并对候选区域进行人脸检测,排除非人脸肤色区域,实现对不同角度人脸的正确检测。该方法可同时解决肤色分割方法对复杂背景图像的高误检率与AdaBoost算法对多姿态人脸图像的低检测率的问题。仿真实验表明,该方法具有检测率高、误检率低、适应性好及鲁棒性强的优点,对具有多姿态、多人脸和复杂背景信息的图像具有较好的检测效果,实用性得到增强。

This paper proposed a face detection algorithm combined skin color segmentation with AdaBoost algorithm. The method utilizes clustering characteristic of skin color and builds Gaussian color model in the YCbCr color space, screening the skin color regions by morphological processing. Then the AdaBoost algorithm is used to train the weak classifier and a strong classifier is formed. The strong classifier is combined to form a cascade classifier and the candidate regions can be detected by AdaBoost cascade classifier, non-face regions in skin color regions can he excluded and then the correct detection of different angles of the faces is achieved. This method can solve the problem of high false detection rate of skin color segmentation with complex background color images and the problem of low detection rate of multi pose images based on the method of AdaBoost algorithm. Simulation experiments show that this method has high detection rate, low false detection rate, good adaptability and strong robustness. It has better detection effect for multi pose and multi-face images and the images of the complex background information, and the practicability is enhanced.