Research‎ > ‎

Imaging Through Turbulence


Images of distant scenes, common in ground-based surveillance and astronomy, are often corrupted by atmospheric turbulence. When recorded by a video, an object in a distance looks distorted due the atmospheric changes between the camera and the object. 

We believe that the main challenge in dealing with atmospheric turbulence is the temporal undersampling that causes seemingly random temporal oscillations and blurring in each video frame. Our main objective is to stabilize these oscillations in both space and time. We propose to apply video frame sharpening and temporal di.usion at the same time. We apply a Sobolev gradient method [1] to sharpen individual frames and mitigate the temporal distortions by the Laplace operator. This eliminates explicit registration that can be computationally expensive.

Furthermore, we use the reconstructed video to construct the latent image when the camera is stationary and the scene is static. We apply an approach related to the lucky-region method but with a different quality criterion to reconstruct a even sharper and more accurate image.

Please refer to [2] for more details & Matlab code can be downloaded at here!




Target board---original

Moving car---original





Target board---SOB+LAP

Moving car---SOB+LAP

Comparison---moving car