计算机工程与应用 ›› 2020, Vol. 56 ›› Issue (1): 203-208.DOI: 10.3778/j.issn.1002-8331.1810-0035

• 图形图像处理 • 上一篇    下一篇

复杂场景下基于R-FCN的小人脸检测研究

李静,降爱莲   

  1. 太原理工大学 信息与计算机学院,山西 晋中 030600
  • 出版日期:2020-01-01 发布日期:2020-01-02

Research on Small Face Detection Based on R-FCN in Complex Scenes

LI Jing, JIANG Ailian   

  1. College of Information and Computer, Taiyuan University of Technology, Jinzhong, Shanxi 030600, China
  • Online:2020-01-01 Published:2020-01-02

摘要: 在复杂场景中准确检测出小的、模糊的和部分遮挡的人脸,仍是人脸检测算法存在的问题。为此,提出基于区域的全卷积网络R-FCN的人脸检测算法,来解决其中的小人脸检测问题。采用完全卷积残差网络ResNet作为主干网络,融合多种新技术,主要包括Squeeze-and-Excitation模块、残差注意力机制等,以提高最终的输出精度。在最具挑战性的人脸检测基准Widerface数据集上测试,结果表明该算法在复杂场景下具有出色的人脸检测效果,对部分遮挡,模糊、人脸姿态变化也具有一定鲁棒性。

关键词: 人脸检测, 区域全卷积神经网络, 残差网络, 复杂场景

Abstract: Accurate detection of small, blurred and partially occluded faces in complex scenes is still a problem with face detection algorithms. To this end, this paper proposes a face-detection algorithm based on the region-based fully convolutional network R-FCN to solve the small face detection problem. The complete convolution residual network ResNet is used as the backbone network, and a variety of new technologies are integrated, including the Squeeze-and-Excitation module and the residual attention mechanism to improve the final output accuracy. Tested on the most challenging face detection benchmark Widerface dataset, the results show that the proposed algorithm has excellent face detection effect in complex scenes, and it is also robust to partial occlusion, blur and face pose changes.

Key words: face detection, region-based fully convolutional network, ResNet, complex scenes