Journal
2022 IEEE 2ND INTERNATIONAL CONFERENCE ON INFORMATION COMMUNICATION AND SOFTWARE ENGINEERING (ICICSE 2022)
Volume -, Issue -, Pages 110-114Publisher
IEEE
DOI: 10.1109/ICICSE55337.2022.9829002
Keywords
Blind image deblurring; generative adversarial networks (GANs); image restoration; deep learning
Funding
- Science and Technology Project of China Huaneng Group Co., LTD. [HKJ20-H88]
Ask authors/readers for more resources
This paper proposes a GAN-based approach, called DeblurFusedGAN (DFGAN), for single-image blind motion deblurring. It improves the performance for image deblurring task by combining attention mechanism and gradient-based filters in the generator network.
Recovering a latent sharp image from a spatially variant blurred image is a challenging task in the field of computer vision especially in blind image deblurring, where the source of the blur kernel is unknown and may vary. To remove the intricate motion blur in the images, recently deep learning-based methods perform latent clean image recovery without the need of knowing the blur kernel explicitly. Unlike conventional blind deblurring methods that assume the blur to be spatially invariant across the image. However, simply stacking convolution layers in deep multi-scale networks does not guarantee the complete removal of motion blur in the images and may lead to a poor performance for blind image deblurring task. Thus, we propose a GAN-based approach for single image blind motion deblurring task in an end-to-end manner, for simplicity its called DeblurFusedGAN (DFGAN). The proposed method improves the performance for image deblurring task by fusing the light-weight attention (LSA) mechanism and gradient-based filters in the generator network. Furthermore, we show the sophisticated performance of our proposed approach both qualitatively and quantitatively in comparison with the other state-of-the-art methods.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available