期刊文献+

VGG网络与多特征融合的遮挡人脸检测 被引量:6

Occlusion face detection based on VGG network and multi-feature fusion
下载PDF
导出
摘要 实际生活中的人脸图像大部分带有遮挡,常常导致待检测人脸的关键信息丢失。针对人脸识别过程中由于遮挡所导致的人脸特征难以提取的问题,本文设计了一种基于VGGNet和多特征点融合的遮挡人脸检测算法。该方法以VGG-16框架为特征提取的骨干网络,在传统VGG网络的全连接层输入前增加遮挡处理单元OCC-Net。在该层中首先采用多特征融合的方法,增强网络对人脸特征的提取;然后利用尺度不变特征转换(SIFT)算法,扩大网络中小尺度特征图,得到更为丰富的互补信息,改善了传统VGG网络由于多次卷积、池化操作后所导致的小尺度特征损失严重的问题;最后改进回归框参数以降低损失函数对遮挡区域的敏感度,通过边框回归得到遮挡区域的位置信息,提升了有遮挡情况下的人脸检测精度。实验结果表明,相较于PCANet、Faster RCNN及未添加OCC-Net的传统VGGNet等常用算法,本文算法在常用的FDDB以及RMFD等遮挡数据集上,均能更精确地定位被遮挡的人脸,证实了该算法的有效性和鲁棒性。 Most face images in real life are occluded, which often leads to the loss of key information of the face to be detected. Aiming at the problem of difficulty in extracting facial features due to occlusion in the process of face recognition, this paper designs a occluded face detection algorithm based on VGGNet and multi-feature point fusion. This method uses the VGG-16 framework as the backbone network for feature extraction, and adds the occlusion processing unit Occlusion-Net(OCC-Net) before the input of the fully connected layer of the traditional VGG network. In this layer, the method of multi-feature fusion is first adopted to enhance the network′s extraction of facial features;then the scale-invariant feature transformation(SIFT) algorithm is used to expand the small-scale feature maps in the network to obtain richer complementary information and improve the traditional VGG network has a serious problem of small-scale feature loss caused by multiple convolution and pooling operations;finally, the regression box parameters are improved to reduce the sensitivity of the loss function to the occluded area, and the position information of the occluded area is obtained through border regression. Improved the accuracy of face detection in the presence of occlusion. The experimental results show that compared with commonly used algorithms such as PCANet, Faster RCNN, and traditional VGGNet without OCC-Net, the algorithm in this paper can more accurately locate the occluded face on the commonly used occlusion data sets such as FDDB and RMFD, which confirms the effectiveness and robustness of the algorithm.
作者 何其霖 穆平安 He Qilin;Mu Pingan(School of Optoelectronic Information and Computer Engineering,Shanghai University of Technology,Shanghai 200082,China)
出处 《电子测量技术》 北大核心 2021年第18期150-154,共5页 Electronic Measurement Technology
基金 2021年学位点引导布局与建设培育项目(XWDB2021105)资助。
关键词 深度学习 人脸检测 遮挡识别 特征提取 deep learning face detection occlusion recognition feature extraction
  • 相关文献

参考文献15

二级参考文献81

  • 1隋婧,金伟其.双目立体视觉技术的实现及其进展[J].电子技术应用,2004,30(10):4-6. 被引量:88
  • 2李武军,王崇骏,张炜,陈世福.人脸识别研究综述[J].模式识别与人工智能,2006,19(1):58-66. 被引量:108
  • 3张文超,山世光,张洪明,陈杰,陈熙霖,高文.基于局部Gabor变化直方图序列的人脸描述与识别[J].软件学报,2006,17(12):2508-2517. 被引量:82
  • 4肖甫,吴慧中,肖亮,汤杨.一种光照鲁棒的图像拼接融合算法[J].中国图象图形学报,2007,12(9):1671-1675. 被引量:14
  • 5袁燕.人脸识别研究综述[J].电脑知识与技术,2007(9):1414-1414. 被引量:4
  • 6Wright J,Yang A Y,Ganesh A. Robust face recognition via sparse representation[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2009,(02):210-227.doi:10.1109/TPAMI.2008.79.
  • 7Yang M,Zhang L,Yang J. Metaface learning for sparse representation based face recognition[A].Hong Kong,China:IEEE Signal Processing Society,2010.1601-1604.
  • 8Ramírez I,Sprechmann P,Sapiro G. Classification and clustering via dictionary learning with structured incoherence and shared features[A].California,USA:IEEE Computer Society,2010.3501-3508.
  • 9Zhang Q,Li B X. Discriminative K-SVD for dictionary learning in face recognition[A].California,USA:IEEE Computer Society,2010.2691-2698.
  • 10Mailhé B,Plumbley M D. Dictionary learning with large step gradient descent for sparse representations[A].Tel-Aviv,Israel:Springer,2012.231-238.

共引文献154

同被引文献54

引证文献6

二级引证文献7

相关作者

内容加载中请稍等...

相关机构

内容加载中请稍等...

相关主题

内容加载中请稍等...

浏览历史

内容加载中请稍等...
;
使用帮助 返回顶部