朱虹,李千目,李德强.基于单个卷积神经网络的面部多特征点定位[J].计算机科学,2018,45(4):273-277, 284
基于单个卷积神经网络的面部多特征点定位
Facial Multi-landmarks Localization Based on Single Convolution Neural Network
投稿时间:2017-03-02  修订日期:2017-06-16
DOI:10.11896/j.issn.1002-137X.2018.04.046
中文关键词:  深度学习,卷积神经网络,面部特征点定位,数据扩增,无约束条件
英文关键词:Deep learning,Convolution neural network,Facial landmarks localization,Data augmentation,Unconstrained condition
基金项目:本文受江苏省重大研发计划社会发展项目:大数据驱动的隧道等城市快速路交通违章取证关键技术研究(SBE2017741114)资助
作者单位E-mail
朱虹 南京理工大学计算机科学与工程学院 南京210094  
李千目 南京理工大学计算机科学与工程学院 南京210094 liqianmu@126.com 
李德强 南京理工大学计算机科学与工程学院 南京210094  
摘要点击次数: 249
全文下载次数: 187
中文摘要:
      深度学习在面部特征点定位领域取得了比较显著的效果。然而,由于姿态、光照、表情和遮挡等因素引起的面部图像的复杂多样性,数目较多的面部特征点定位仍然是一个具有挑战性的问题。现有的用于面部特征点定位的深度学习方法是基于级联网络或基于任务约束的深度卷积网络,其不仅复杂,且训练非常困难。为了解决这些问题,提出了一种新的基于单个卷积神经网络的面部多特征点定位方法。与级联网络不同,该网络包含了3组堆叠层,每组由两个卷积层和最大池化层组成。这种网络结构可以提取更多的全局高级特征,能更精确地表达面部特征点。大量的实验表明,所提方法在姿态、光照、表情和遮挡等变化复杂的条件下优于现有的方法。
英文摘要:
      Facial landmarks localization methods using deep learning network technology have achieved prominent effect.However,the localization of larger number of facial landmarks still has lots of challenges due to the complex diversities in face images caused by pose,expression,illumination and occlusion,etc.The existing deep learning methods for face and mark localization are based on cascaded networks or tasks-constrained deep convolutional network(TCDCN),which are complicated and difficult to train.To solve these problems,a new method of facial multi-landmarks location based on single convolution neural network was proposed.Unlike cascaded networks,the network consists of three stacks,and each group consists of two convolutional layers and a max-pooling layer.This network structure can extract more global high-level features,which express the facial landmarks more precisely.Extensive experiments show that the approach outperforms the existing methods in the complex conditions such as pose,illumination,expression and occlusion.
查看全文  查看/发表评论  下载PDF阅读器