基于改进的Adaboost算法的人脸检测系统

摘要

With the development of social informatization and intelligence, face detection technology plays an increasingly important role in business, culture and other fields. The performance requirements of face detection systems for society are also increasing. Open source computer vision library OpenCV implements a number of image processing algorithms, including Adaboost algorithm to train Haar classifier for high-accuracy face detection. The commonly used Haar-Like feature-based AdaBoost face detection algorithm still has some disadvantages, such as high missed detection rate and false detection rate and low detection efficiency. In order to make up these deficiencies, we add a new Haar-Like feature that matches the distribution of human face in the original Haar-Like feature set to increase the detection rate and reduce the false alarm rate. And through the introduction of skin detection technology to screen the human skin area as an alternative area, remove most of the non-skin color areas in the picture, in order to improve detection efficiency. Through testing and performance comparison, the improved face detection system has better accuracy and higher detection efficiency.%随着社会的信息化与智能化发展,人脸检测技术在商业、文化等领域扮演着日益重要的角色,社会对人脸检测系统的性能要求也越来越高.开源计算机视觉库OpenCV中实现了众多图像处理算法,其中包括使用Adaboost算法训练出Haar分类器,用以进行高准确率人脸检测.常用的基于Haar-Like特征的AdaBoost人脸检测算法还存在着不足之处,例如漏检率和误检率较高,检测效率较低等.针对这些不足之处,在原有的Haar-Like特征集中加入新的符合人脸器官分布的Haar-Like特征,以提高检测率,降低虚警率.并通过引进肤色检测技术来筛选人体肤色区域作为备选区域,剔除图片中大部分的非肤色区域,以提高检测效率.通过测试与性能对比,改进的人脸检测系统具有更好的准确率和更高的检测效率.