What is it about?
High dynamic range (HDR) imaging by combining multiple low dynamic range (LDR) images of different exposures provides a promising way to produce high quality photographs. However, the misalignment between the input images leads to ghosting artifacts in the reconstructed HDR image. In this paper, we propose a novel neural network that merges and fuses the LDR inputs, processing the data using multiple frequency representations.
Featured Image
Photo by op23 on Unsplash
Why is it important?
We present a novel spatial attention mechanism and transform domain processing to reduce ghosting artifacts.
Perspectives
This article takes a different approach to HDR imaging in that it does not apply an optical flow step to try to align pixels, instead relying on a novel spatial attention mechanism, resulting in an efficient algorithm. The cross domain processing is shown to give a measurable improvement in performance.
Greg Slabaugh
Queen Mary University of London
Read the Original
This page is a summary of: DomainPlus: Cross Transform Domain Learning towards High Dynamic Range Imaging, October 2022, ACM (Association for Computing Machinery),
DOI: 10.1145/3503161.3547823.
You can read the full text:
Contributors
The following have contributed to this page