基于MB_LBP旋转不变特征的AdaBoost人脸检测算法研究

摘要

传统的基于Haar特征的AdaBoost人脸检测算法,由于Haar特征数量过多,导致训练时间过久,而且不能快速检测出人脸.针对这一问题,本文提出一种基于多块局部二值模式(Multi-block Local Binary Pattern,MB_LBP)特征的AdaBoost人脸检测算法,这种MB_LBP特征结合了旋转不变局部二值模式(Local Binary Patterns,LBP)描述符,表达能力更强,特征数量更少.仿真结果表明,在训练时间大幅缩减的同时,使用MB_LBP特征时可以达到Haar特征的检测效果,且检测速度大大提高.%The training time of the traditional AdaBoost face detection algorithm based on the Haar features is too long,and the face can not be detected quickly due to the excessive number of Haar features.Aiming at this problem,the AdaBoost face detection algorithm based on the multi-block local binary pattern (MB_LBP) features is proposed in this paper.The MB_LBP features are combined with the rotation invariant local binary pattern (LBP) descriptor,which have stronger representation ability and fewer features.The simulation results show that the MB_LBP features performed as well as the Haar features do,and the training time is significantly reduced at the same time,and the detection rate is greatly improved.