Blind Image Deblurring: a Review

Abstract

In this paper, we first formulate the blind image deblurring problem and explain why it is challenging. Next, we bring some psychological and cognitive studies on the way our human vision system deblurs. Then, relying on several previous reviews, we discuss the topic of metrics and datasets, which is non-trivial to blind deblurring, and introduce some typical optimization-based methods and learning-based methods. Finally, we review the highlights of the theoretical analysis part of Levin et al. [32], which could answer some very fundamental questions about blind deblurring.

Enhanced Mask R-CNN blur instance segmentation based on generated adversarial network

“…Image blind deblurring is not only a basic problem in the field of low-level vision but also a frontier topic in the field of computational imaging. It is assumed that the blurred image g is generated by the convolution of the original clear image u and the space-invariant linear blur kernel k, and may contain additive white Gaussian noise n. The mathematical expressions [1] is as follows:…”

Bilevel inverse problems in neuromorphic imaging

“…Image deblurring is a classic example of an inverse problem, where the goal is to reconstruct the unknown original image from a degraded, blurred observation. In image deblurring, the degradation process can be modeled as an interaction between the original image and a blur kernel which represents the blur caused by various factors such as movement [21,22].…”

“…In the context of image deblurring, the field can be divided into two categories, blind and non-blind, based on the amount of information known about the blur kernel [1,16,18,20,21]. In the non-blind case, the blur kernel is known which leads to a simpler albeit ill-posed inverse problem [18,20].…”

“…In the non-blind case, the blur kernel is known which leads to a simpler albeit ill-posed inverse problem [18,20]. In the blind case, no information about the blur kernel is available, making the deblurring problem more challenging as the blur kernel must be estimated from the degraded image in addition to the original image [16,21]. The semi-blind case is a particular instance of a blind image deblurring problem in that some but not all information about the blur kernel is available [6,13,15].…”