English

SR-R$^2$KAC: Improving Single Image Defocus Deblurring

Computer Vision and Pattern Recognition 2023-08-01 v1

Abstract

We propose an efficient deep learning method for single image defocus deblurring (SIDD) by further exploring inverse kernel properties. Although the current inverse kernel method, i.e., kernel-sharing parallel atrous convolution (KPAC), can address spatially varying defocus blurs, it has difficulty in handling large blurs of this kind. To tackle this issue, we propose a Residual and Recursive Kernel-sharing Atrous Convolution (R2^2KAC). R2^2KAC builds on a significant observation of inverse kernels, that is, successive use of inverse-kernel-based deconvolutions with fixed size helps remove unexpected large blurs but produces ringing artifacts. Specifically, on top of kernel-sharing atrous convolutions used to simulate multi-scale inverse kernels, R2^2KAC applies atrous convolutions recursively to simulate a large inverse kernel. Specifically, on top of kernel-sharing atrous convolutions, R2^2KAC stacks atrous convolutions recursively to simulate a large inverse kernel. To further alleviate the contingent effect of recursive stacking, i.e., ringing artifacts, we add identity shortcuts between atrous convolutions to simulate residual deconvolutions. Lastly, a scale recurrent module is embedded in the R2^2KAC network, leading to SR-R2^2KAC, so that multi-scale information from coarse to fine is exploited to progressively remove the spatially varying defocus blurs. Extensive experimental results show that our method achieves the state-of-the-art performance.

Keywords

Cite

@article{arxiv.2307.16242,
  title  = {SR-R$^2$KAC: Improving Single Image Defocus Deblurring},
  author = {Peng Tang and Zhiqiang Xu and Pengfei Wei and Xiaobin Hu and Peilin Zhao and Xin Cao and Chunlai Zhou and Tobias Lasser},
  journal= {arXiv preprint arXiv:2307.16242},
  year   = {2023}
}

Comments

Submitted to IEEE Transactions on Cybernetics on 2023-July-24

R2 v1 2026-06-28T11:43:49.378Z