基于子模式下LBP-HOG特征融合的单样本人脸识别方法

A single sample face recognition method based on LBP-HOG feature fusion in submode

针对目前单样本人脸识别率不高这一问题,本文提出一种基于子模式下的分层LBP和金字塔模式HOG特征相融合的单样本人脸识别方法。该方法针对人脸不同部位对人脸识别所做贡献的程度,先通过已有人脸不同部位的分类器将人脸不同器官提取出来并以此为基准将其分为不同的子图像。然后针对LBP描述子提取的纹理特征数量较少且不能很好的描述图像边缘和方向信息等问题,将分层LBP与金字塔不同层级的HOG特征相融合的方法作用在每一个子图像上,得到每一个子图像的融合特征向量,计算训练样本与测试样本对应的每一个子图像的融合特征的欧氏距离并且乘上预先设定的该子图像对应的权重参数,然后将它们相加得到最终的欧氏距离,通过阈值判断该人脸所属id。最后通过在ORL人脸库上进行实验,结果表明本文提出的方法比现有单样本人脸识别方法识别率更高。

Aimed at the problem of single sample face recognition rate is not high,In order to improve the recognition rate of single sample face recognition,this paper proposes a single sample face recognition method based on the combination of layered LBP and HOG features of pyramid mode under submode.According to the contribution of different parts of face to face recognition,this method firstly extracts different parts of face through the existing classifier of different parts of face and divides them into different sub-images.Then,aiming at the problem that the number of texture features extracted by LBP descriptors is small and the image edge and orientation information cannot be well described,the method of fusion of layered LBP and HOG features at different levels of the pyramid is applied to each sub-image to obtain the fusion feature vector of each sub-image,The Euclidean distance between the training sample and the fusion features of each sub-image corresponding to the test sample is calculated and multiplied by the pre-set weight parameters of the sub-image,then the final Euclidean distance is obtained by adding them together,and the id of the face is determined by the threshold.Finally,the experiment on ORL face database shows that the proposed method has a higher recognition rate than the existing single sample face recognition method,The recognition rate of ORL face database is up to 89%.