Realistic Blur Synthesis for Learning Image Deblurring

TL;DR

This paper introduces RSBlur, a dataset of real blurred images with sharp sequences, and a new synthesis pipeline that produces more realistic blur for training deblurring models, improving their performance on real images.

Contribution

The paper presents RSBlur, a dataset with real blurred images and a novel blur synthesis pipeline that enhances the realism of synthetic data for deblurring.

Findings

The dataset enables detailed analysis of real vs. synthetic blur differences.

The new synthesis pipeline produces more realistic blur effects.

Models trained with the pipeline perform better on real blurred images.

Abstract

Training learning-based deblurring methods demands a tremendous amount of blurred and sharp image pairs. Unfortunately, existing synthetic datasets are not realistic enough, and deblurring models trained on them cannot handle real blurred images effectively. While real datasets have recently been proposed, they provide limited diversity of scenes and camera settings, and capturing real datasets for diverse settings is still challenging. To resolve this, this paper analyzes various factors that introduce differences between real and synthetic blurred images. To this end, we present RSBlur, a novel dataset with real blurred images and the corresponding sharp image sequences to enable a detailed analysis of the difference between real and synthetic blur. With the dataset, we reveal the effects of different factors in the blur generation process. Based on the analysis, we also present a novel blur synthesis pipeline to synthesize more realistic blur. We show that our synthesis pipeline can improve the deblurring performance on real blurred images.