Use fast R-CNN and cascade structure for face detection

Face detection is challenging in real-world due to various poses, occlusions, lighting conditions and so on. Recently, deep learning achieves excellent performance on this task. In this paper, we propose a novel deep cascade convolutional network that uses Fast Region-based Convolutional Network (Fa...

Full description

Saved in:
Bibliographic Details
Published in2016 Visual Communications and Image Processing (VCIP) pp. 1 - 4
Main Authors Kai Wang, Yuan Dong, Hongliang Bai, Yu Zhao, Kun Hu
Format Conference Proceeding
LanguageEnglish
Published IEEE 01.11.2016
Subjects
Calibration
cascade
Face
Face detection
Fast-RCNN
FDDB
Feature extraction
Proposals
state-of-the-art
Training
Training data
Online AccessGet full text
DOI10.1109/VCIP.2016.7805472

Cover

More Information
Summary:Face detection is challenging in real-world due to various poses, occlusions, lighting conditions and so on. Recently, deep learning achieves excellent performance on this task. In this paper, we propose a novel deep cascade convolutional network that uses Fast Region-based Convolutional Network (Fast R-CNN) [3] at the end of the structure. Our method achieves outstanding performance on the FDDB [1] and AFW [2] datasets. Especially, it achieves a high recall of 91.87% on the challenging FDDB benchmark, outperforming the state-of-the-art methods. We adopt a cascaded structure to quickly reject the background regions first. At the last stage, we choose Fast R-CNN to deal with the challenging candidates and it performs well.
DOI:10.1109/VCIP.2016.7805472