作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程 ›› 2021, Vol. 47 ›› Issue (11): 276-282. doi: 10.19678/j.issn.1000-3428.0059379

• 开发研究与工程应用 • 上一篇    下一篇

基于深度残差网络和注意力机制的人脸检测算法

陶施帆, 李玉峰, 黄煜峰, 蓝晓宇   

  1. 沈阳航空航天大学 电子信息工程学院, 沈阳 110136
  • 收稿日期:2020-08-27 修回日期:2020-10-26 发布日期:2020-11-16
  • 作者简介:陶施帆(1995-),女,硕士研究生,主研方向为人脸检测与识别;李玉峰,教授、博士;黄煜峰,讲师、博士;蓝晓宇,副教授、博士。
  • 基金资助:
    国家高分专项辽宁湿地遥感监测与生态旅游遥感调查产业化应用项目(70-Y40-G09-9001-18/20);辽宁省自然科学基金(20180550334);辽宁省教育厅项目(L201701);国家青年科学基金(61801308);辽宁省“兴辽英才计划”项目(XLYC1907195)。

Face Detection Algorithm Based on Deep Residual Network and Attention Mechanism

TAO Shifan, LI Yufeng, HUANG Yufeng, LAN Xiaoyu   

  1. School of Electronic Information Engineering, Shenyang University of Aeronautics and Astronautics, Shenyang 110136, China
  • Received:2020-08-27 Revised:2020-10-26 Published:2020-11-16

摘要: 人脸检测技术作为一种人员身份识别的主流技术被广泛应用于人们的日常生活中。然而在特定应用场景中,当人脸被遮挡或人脸目标非常密集时,人脸识别的检测性能急剧下降。提出一种基于深度残差网络和注意力机制的高精度人脸检测算法。使用残差网络ResNet-50并结合IoU损失函数提高人脸检测精度,并利用注意力机制优化突出脸部区域特征,在此基础上采用非极大值抑制方法增强算法鲁棒性。在公开FDDB数据集上的实验结果表明,该算法的准确率达到96.1%相比传统卷积网络VGG-16算法提高1.6个百分点。

关键词: 人脸检测, 非极大值抑制, 注意力机制, 残差网络, IoU损失函数

Abstract: As a mainstream technology for personal identification,face detection has been widely used in daily life. However,in some application scenarios,the face recognition performance will be decreased dramatically when the face is occluded or the face targets are very dense.To address the problem,a high-precision face detection algorithm based on deep residual network and attention mechanism is proposed.The algorithm utilizes the residual network ResNet-50 combined with the IoU loss function to improve the accuracy of face detection,and then the attention mechanism is employed to optimize the prominent facial area features.On this basis,the Non-Maximum Suppression(NMS) method is used to enhance the robustness of the algorithm.The experimental results on the public dataset,FDDB,show that the accuracy rate of the proposed algorithm reaches 96.1%,which is 1.6% higher than that of the traditional algorithm based on VGG-16.

Key words: face detection, Non-Maximum Suppression(NMS), attention mechanism, residual network, IoU loss function

中图分类号: