What problem does this paper attempt to address?

The problem that this paper attempts to solve is to mitigate the impact of atmospheric turbulence on images and videos by evaluating different neural network architectures. Specifically, the authors aim to: 1. **Study the applicability of existing neural network architectures**: Evaluate the effectiveness of existing neural network architectures in dealing with image blurring problems caused by atmospheric turbulence, and determine whether these architectures can be directly applied to the correction of atmospheric turbulence. 2. **Create a publicly available dataset**: To promote research in this field, the authors created a publicly available dataset named SOTIS (Simulated Open Turbulence Image Set), which contains image sequences under the influence of simulated atmospheric turbulence and their corresponding clear images (i.e., ground - truth). This helps researchers conduct quantitative evaluations and model training. 3. **Propose an evaluation method**: Given the diversity of atmospheric turbulence correction algorithms, the authors proposed a unified evaluation method to compare the performance of different algorithms in a fair and quantitative manner. They used Peak Signal - to - Noise Ratio (PSNR) and Structural Similarity Index (SSIM) as the main evaluation indicators. 4. **Explore the potential of deep learning in atmospheric turbulence correction**: By comparing classical algorithms and deep - learning - based algorithms, the authors hope to explore the performance of neural networks in this specific task and provide guidance for future research. ### Specific problem description Atmospheric turbulence can cause inhomogeneity in the refractive index in the optical path, resulting in image blurring and geometric distortion. Traditional correction methods are usually divided into two steps: - **Stabilization**: Compensate for geometric deformations in the image. - **Deblurring**: Restore the blurred image. However, due to the anisoplanatic nature of the blurring caused by atmospheric turbulence, the effectiveness of traditional methods is limited. Therefore, the authors attempt to use neural networks to solve this problem, and pay special attention to the following aspects: - **Reusability of existing neural networks**: Test whether existing neural network architectures for deblurring can be directly applied to atmospheric turbulence correction. - **End - to - end training**: Explore whether the process can be simplified by end - to - end training and the stabilization step can be removed. ### Main contributions 1. **SOTIS dataset**: Provide a large number of image sequences under the influence of simulated atmospheric turbulence and their corresponding clear images, filling the gap in the lack of publicly available datasets in this field. 2. **Systematic evaluation**: Through the systematic evaluation of multiple neural network architectures, reveal the performance differences of different architectures in the atmospheric turbulence correction task. 3. **Application prospects of deep learning**: Demonstrate the potential of deep learning in atmospheric turbulence correction and provide a basis for further research. Through these efforts, the authors hope to promote the development of atmospheric turbulence correction technology, especially to improve image quality in long - distance sensor applications.

Evaluation of neural network algorithms for atmospheric turbulence mitigation

Ground-to-air Aircraft Infrared Image Deblurring Based on Imaging Degradation Simulation

Effective Framework for Space Target Detection Through Atmospheric Turbulence

NeRT: Implicit Neural Representations for General Unsupervised Turbulence Mitigation

Atmospheric Turbulence Removal with Complex-Valued Convolutional Neural Network

Deep Learning Techniques for Atmospheric Turbulence Removal: A Review

A comparison of different atmospheric turbulence simulation methods for image restoration

Imaging Through the Atmosphere Using Turbulence Mitigation Transformer

Learning an Occlusion-Aware Network for Video Deblurring

Joint Generative Image Deblurring Aided by Edge Attention Prior and Dynamic Kernel Selection

Spatio-Temporal Turbulence Mitigation: A Translational Perspective

A new approach for atmospheric turbulence removal using low-rank matrix factorization

Removing Turbulence Effect Via Hybrid Total Variation and Deformation-Guided Kernel Regression

Real-World Atmospheric Turbulence Correction via Domain Adaptation

Energy-conserving neural network for turbulence closure modeling

DanceCam: atmospheric turbulence mitigation in wide-field astronomical images with short-exposure video streams

Examining the Impact of Blur on Recognition by Convolutional Networks

DeTurb: Atmospheric Turbulence Mitigation with Deformable 3D Convolutions and 3D Swin Transformers

Atmospheric Turbulence Removal with Video Sequence Deep Visual Priors

Neutralizing the impact of atmospheric turbulence on complex scene imaging via deep learning