Image Processing Algorithms for Imaging Through Atmospheric Turbulence

Abstract

In this grant, we aimed at studying image processing algorithms for turbulence mitigation related problems. This grant was the continuation of our previous grant (FA9550-15-1-0065). We developed an original approach, called BATUD (Blind Atmospheric TUrbulence Deconvolution), to perform atmospheric deblurring. We created two datasets, OTIS and SOTIS, that we made publicly available. The first one is a small dataset containing real static and dynamic (i.e. with a moving target) sequences. The second one is a very large dataset created thanks to an atmospheric turbulence simulator. The purpose of such dataset is twofold: 1) to run extensive algorithm performance evaluations (we have run the evaluation of unsupervised algorithms during this project); 2) to develop dedicated neural network based techniques in a near future. We also investigated the use of empirical wavelets to perform deblurring tasks which led to a self-adapting algorithm.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
May 25, 2022
Accession Number
AD1230270

Entities

People

  • Jérôme Gilles

Organizations

  • Salk Institute for Biological Studies

Tags

Fields of Study

  • Computer science

Readers

  • Neural Network Machine Learning.
  • Ocean-Atmosphere Mesoscale Modeling, Data Assimilation, and Flux Boundary Layers
  • Research Science/Academic Research

Technology Areas

  • AI & ML
  • AI & ML - Neural Networks