首页> 外文会议>2017 International Conference on Security, Pattern Analysis, and Cybernetics >Missing area completion in facial images using maximum-correntropy-criterion regularized cascading autoencoder
【24h】

Missing area completion in facial images using maximum-correntropy-criterion regularized cascading autoencoder

机译:使用最大熵准则正则级联自编码器的面部图像中的缺失区域补全

获取原文
获取原文并翻译 | 示例

摘要

Missing area completion for human facial image plays a significant role in both academic area and people's daily life. The completion performance using neural network relies heavily on both architectural design and adopted cost functions. To this end, this paper proposed a maximum-correntropy-criterion regularized cascading autoencoder for facial image completion. The results of experiments in LFWcrop Face Dataset and Frey Face Database show that the MCC regularized CAE network is well performed to generate general human facial features in the missing area. The completion accuracy improves around 5% compared to using Restricted Boltzmann Machine (RBM) in the same dataset.
机译:人脸图像的缺失区域补全在学术领域和人们的日常生活中都起着重要作用。使用神经网络的完井性能在很大程度上取决于体系结构设计和采用的成本函数。为此,本文提出了一种用于脸部图像完成的最大熵准则正则化级联自动编码器。 LFWcrop人脸数据集和Frey人脸数据库中的实验结果表明,MCC规范化的CAE网络表现良好,可以在缺失区域生成一般的人脸特征。与在同一数据集中使用受限玻尔兹曼机(RBM)相比,完成精度提高了5%左右。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱:kefu@zhangqiaokeyan.com

京公网安备:11010802029741号 ICP备案号:京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有
  • 客服微信

  • 服务号