Edge Detection of Motion-Blurred Images based on GANs
Feng Li (Donghua university), Jiyu Li (Donghua University)*
The 33rd British Machine Vision Conference
Abstract
Motion blur is a challenging problem in many image processing tasks. It leads to the degradation of the image, especially the edge information within. This paper introduces an edge detection algorithm for motion-blurred images based on Generative Adversarial Networks, treating edge detection as an image translation problem. We determined the components and parameters of the networks through experimental comparisons and finally adopted U-Net and PatchGAN as the backbones. We also proposed a new loss calculation method called Motion Loss, which tolerates the random offset of edges due to motion blur. Finally, we performed several experiments on the GOPRO dataset. The results showed that applying Motion Loss could lead to better edge results and that our method worked well in the edge detection of motion-blurred images.Video
Citation
@inproceedings{Li_2022_BMVC,
author = {Feng Li and Jiyu Li},
title = {Edge Detection of Motion-Blurred Images based on GANs},
booktitle = {33rd British Machine Vision Conference 2022, {BMVC} 2022, London, UK, November 21-24, 2022},
publisher = {{BMVA} Press},
year = {2022},
url = {https://bmvc2022.mpi-inf.mpg.de/0266.pdf}
}
Copyright © 2022
The British Machine Vision Conference is organised by