Blind Image Deblurring with FFT-ReLU Sparsity Prior

TL;DR

This work introduces a method that leverages a prior which targets the blur kernel to achieve effective deblur-ring across a wide range of image types, and offers up to two times faster inference, making it a highly efficient solution.

Abstract

Blind image deblurring is the process of recovering a sharp image from a blurred one without prior knowledge about the blur kernel. It is a small data problem, since the key challenge lies in estimating the unknown degrees of blur from a single image or limited data, instead of learning from large datasets. The solution depends heavily on developing algorithms that effectively model the image degradation process. We introduce a method that leverages a prior which targets the blur kernel to achieve effective deblurring across a wide range of image types. In our extensive empirical analysis, our algorithm achieves results that are competitive with the state-of-the-art blind image deblurring algorithms, and it offers up to two times faster inference, making it a highly efficient solution.

Figures (6)

• Figure 1: Results of our blind image deblurring algorithm, compared with other state-of-the-art algorithms. From left to right: 1) Input blurry image, followed by results from 2) Chen et al.Chen 3) Wen et al.PMP 4) Pan et al.PanText, 5) Pan et al.DCP and 6) our algorithm.
• Figure 2: Decreased Sparsity in Blurry Images after RFT, as demonstrated in images from Sun et al.SunDataset, TextOCR dataset, singh2021textocr LOL dataset, loldataset and LFW dataset LFWTech.
• Figure 3: Using gradient descent, FI and RFT(I) converge reasonably within 100 steps.
• Figure 4: PSNR comparison for Köhler et al. dataset
• Figure 5: Results of our blind image deblurring algorithm, compared with other state-of-the-art algorithms. From left to right: 1) Input blurry image, followed by results from 2) Chen et al.Chen 3) Wen et al.PMP 4) Pan et al.PanText, 5) Pan et al.DCP and 6) our algorithm.