首页> 外文会议>International Conference on Computer Science and Network Technology >Face detection based on multi task learning and multi layer feature fusion
【24h】

Face detection based on multi task learning and multi layer feature fusion

机译:基于多任务学习的面部检测和多层特征融合

获取原文

摘要

Face detection and facial feature location are two key parts of face recognition system. Usually, these two links are treated as two separate tasks, ignoring the correlation between tasks. In addition, most of the face detection algorithms based on deep convolution neural networks focus only on high-level semantic information of the image, and do not take full advantage of the underlying details of the image. In order to further improve the performance of face detection, we propose a face detection algorithm based on multi task learning and multilayer feature fusion. The proposed method integrates three tasks, namely, face classification, facial feature location, and bounding box regression, into a framework that takes full advantage of the correlation between multiple tasks and performs simultaneous learning over multiple tasks. At the same time, in order to make full use of the low-level details and high-level semantic information of the image, multi layer feature fusion technology is adopted. Finally, we test it on the face detection evaluation database FDDB. Experimental results show that the proposed algorithm has good performance in face detection.
机译:面部检测和面部特征位置是面部识别系统的两个关键部分。通常,这两个链接被视为两个单独的任务,忽略任务之间的相关性。此外,基于深度卷积神经网络的大多数面部检测算法仅关注图像的高电平语义信息,并且不充分利用图像的底层细节。为了进一步提高面部检测的性能,我们提出了一种基于多任务学习和多层特征融合的面部检测算法。该方法将三个任务,即面部分类,面部特征位置和边界框回归集成到一个框架中,该框架充分利用多个任务之间的相关性并执行多个任务的同时学习。同时,为了充分利用图像的低级细节和高电平语义信息,采用多层特征融合技术。最后,我们在面部检测评估数据库FDDB上测试它。实验结果表明,该算法在面部检测方面具有良好的性能。

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