摘要
提出了具有鲁棒性的基于局部二值模式LBP纹理提取人脸特征点的主动外观模型AAM算法(LBP-AAM).首先建立3种人脸模型实例(正面,左转及右转人脸模型);然后应用LBP判断和预测测试图片的旋转类型,依据预测结果选择合理的模型实例去匹配;最后提取出人脸特征点.实验结果证明本方法比传统的AAM方法在精度上提高了27%,效率上提高了9%.
In this paper,we propose a robust facial feature points extraction method-LBP-AAM（L-AAM）,using active appearance model（AAM） based on local binary pattern（LBP） texture features.We firstly generated three types of model instances（frontal,left-rotated and right-rotated）,and LBP was used to judge the type of test facial image and predict the rotation of the test the face,and according to the prediction we selected proper model instances as the fitting model.Finally we extracted the feature points of the face.Experimental results proved that this method increased the fitting accuracy rate by about 27% and the time consumption was decreased by about 9% comparing with the standard AAM method.
出处
《应用科技》
CAS
2011年第4期35-38,共4页
Applied Science and Technology
