Related Experiment Video
Updated: Oct 12, 2025
10:16
Digital Inline Holographic Microscopy DIHM of Weakly-scattering Subjects
Published on: February 8, 2014
12.4K
Image restoration for real-world under-display imaging.
Optics Express
|November 23, 2021
Summary
We developed a new method and synthetic dataset (CalibPSF) to improve under-display (UD) imaging. Our two-stage neural network effectively restores images degraded by UD optical systems, outperforming existing techniques.
Area of Science:
- Computer Vision
- Image Processing
- Optical Engineering
Background:
- Under-display imaging techniques aim to increase screen-to-body ratios in full-screen devices.
- Current image restoration algorithms struggle with real-world under-display images, particularly those with strong light sources.
Purpose of the Study:
- To address the limitations of existing methods for under-display image restoration.
- To introduce a novel synthetic dataset and a two-stage neural network for improved under-display imaging.
Main Methods:
- Generation of the CalibPSF dataset using calibrated high dynamic range point spread functions (PSFs) of under-display optical systems.
- Development of a two-stage neural network to sequentially address color distortion and diffraction degradation.
- Evaluation of the proposed method on a captured real-world test set.
Main Results:
- The proposed method demonstrates superior performance in restoring under-display images across various dynamic range scenes.
- The synthetic CalibPSF dataset facilitates effective training for under-display image restoration.
- The two-stage network successfully mitigates color distortion and diffraction artifacts.
Conclusions:
- The novel approach significantly enhances the quality of images captured through under-display imaging systems.
- This work provides a robust solution for under-display image restoration challenges, especially in high dynamic range scenarios.
