Explore Image Deblurring via Blur Kernel Space

TL;DR

This paper presents a novel approach to blind image deblurring by encoding blur operators into a learnable kernel space, enabling the handling of unseen blurs and integration with deep learning models.

Contribution

The paper introduces a differentiable kernel space encoding for blur operators, allowing blind deblurring of unseen kernels and domain-specific blur synthesis, which is a novel advancement over existing methods.

Findings

Effective blind deblurring of unseen kernels

Ability to transfer blur operators across domains

Compatibility with deep neural network models

Abstract

This paper introduces a method to encode the blur operators of an arbitrary dataset of sharp-blur image pairs into a blur kernel space. Assuming the encoded kernel space is close enough to in-the-wild blur operators, we propose an alternating optimization algorithm for blind image deblurring. It approximates an unseen blur operator by a kernel in the encoded space and searches for the corresponding sharp image. Unlike recent deep-learning-based methods, our system can handle unseen blur kernel, while avoiding using complicated handcrafted priors on the blur operator often found in classical methods. Due to the method's design, the encoded kernel space is fully differentiable, thus can be easily adopted in deep neural network models. Moreover, our method can be used for blur synthesis by transferring existing blur operators from a given dataset into a new domain. Finally, we provide experimental results to confirm the effectiveness of the proposed method.