摘要
针对纯粹引入Haar-like特征的人脸检测算法误检率高且受光照条件影响较大的问题,在保证实时性基本不受影响的前提下,以提高检测精度及增强对光照的鲁棒性为目的,提出了基于Adaboost算法结合MB-LBP算子和联合积分直方图(JIH)及Haar模板进行人脸特征提取的新方法。测试结果验证了新方法训练的分类器比单纯使用Haar特征或MB-LBP效果好,证明了提出的方法的有效性。
To solve the problems of high false detection rate and the great influence of the lighting variance in traditional face detection algorithm,Adaboost algorithm for face detection based on multi-block local binary pattern（MB-LBP）features,Haar templates and joint integral histogram（JIH）is proposed.Which can not affect the real-time performance,improve the detection accuracy and enhance robustness of light.Test results demonstrate that the classifier is trained by the new method is better than only by Haar feature or MB-LBP.Experimental results prove the effectiveness of the proposed method.
出处
《计算机与数字工程》
2016年第5期944-947,共4页
Computer & Digital Engineering
