The FDR-LBP (Principal Direction Rotation Local Binary Pattern) features are presented to describe rotation face in image plane .We add rotation offset value for forming rotation LBP features based on arbitrary neighborhood LBP feature and then train a rotation face classifier by Adaboost algorithm .We use the classifier to detect the region of rotation face and then verify the result .A new principal direction algorithm is presented for calculating the angle of face rotation ,with the angle to normalize the rotation face .The experiment results indicate that the new method can detect faces under all degree rotation in image plane with high speed ,the detection rate is 94% and the normalization difference is below 6 degree .It can satisfy face detection system requirements of all degree rotation detection ,with high detection rate and low normalization difference .%提出了主方向旋转LBP特征,对图像中的平面旋转人脸特征进行描述。以任意邻域LBP特征为基础,加入旋转角度偏移值构成旋转LBP特征,通过Adaboost算法训练出旋转人脸分类器,应用旋转人脸分类器检测图像中可能包含旋转人脸的区域,并对结果进行验证。为了提高扶正精度,在旋转LBP特征的基础上加入主方向值,并提出旋转LBP特征的主方向计算方法,有效的提高了扶正精度。经实验证明,新方法能够以较快速度检测所有角度的平面旋转人脸,正确检测率为94%,角度误差在6度以下。满足平面旋转人脸检测系统对全部角度检测、高检测率、低扶正误差的要求。
展开▼