基于几何特征与新Haar特征的人脸检测算法
糜元根;陈丹驰;季鹏
【期刊名称】《传感器与微系统》 【年(卷),期】2017(036)002
【摘要】针对普通Adaboost算法训练时间长,对复杂背景下(暗光、多角度、多姿态等)人脸检测识别率较低的问题,提出了一种改进的结合肤色检测及几何特征的Adaboost算法.采用肤色粗检筛选出候选人脸区域,同时采用新的非对称的Haar特征来训练分类器,进一步加强检测性能,提升鲁棒性和复杂背景下的宽容度实验将此算法应用到一个嵌入式系统中,结果表明:在各种复杂背景下的人脸检测中鲁棒性和宽容度均提升很多,误识率进一步降低,并且在嵌入式人脸检测的系统中具有很好的可移植性和实用性.%Aiming at problem that training time of ordinary Adaboost algorithm is long and under complex background such as dark light,multi angle,multi pose,the recognition rate of face detection is low,propose an improved Adaboost algorithm based on skin color detection and geometric features.The algorithm screens out candidate face regions based on skin color detection,at the same time,a new non-symmetric Haar feature is used to train the classifier,which further enhances detection performance and promote robustness and tolerance in the context of complexity.In the experiment,the algorithm is applied to an embedded system,the results show that the algorithm can improve the robustness and tolerance of face detection in all kinds of complex background,and the error
基于几何特征与新Haar特征的人脸检测算法
