Wide area image restoration and turbulence visualisation

Associate Professor Donald Fraser Dr Andrew Lambert Mr Reza Sayyah (Research Associate) Mr James Webb (PhD Student) Mr Murat Tahtali (PhD Student)

GO DIRECTLY TO MOVIE PAGE FOR SURVEILANCE OR ASTRONOMICAL IMAGING

Introduction

Imaging through a distorting medium such as the atmosphere results in a distorted image. This is because the atmosphere acts to place random variations of refractive index in the path of the light. By taking very short exposures (typically 1 to 10 ms), the motion of the atmosphere can be "frozen", and each exposure is distorted by a random but fixed medium. Processing the sequence of images, we can obtain a more accurate image (with a signal to noise ratio much higher than that of a single exposure), than is possible for long exposure integration.

These techniques have been known and algorithms developed for the restoration, particularly in the field of astronomical imaging, for sometime. However, these apply to the scenario where the telescope viewing angle is smaller than the isoplanatic patch. In these cases, each entire exposure is distorted by a uniform but random Point Spread Funciton (PSF). The atmospheric distortion renders itself as speckle, through multi-path propagation through different "seeing cells". If the viewing angle is larger than the isoplanatic patch then the PSF varies across the exposure, as different parts of the image are derived through different regions of the distorting medium.

The non-uniform PSF problem applies to wide-area image restoration, and is most commonly visible as gemoetric distortion between exposures. By processing these exposures in a sequence, using a heirachical matching algorithm we are able to develop vectors of the shift (relative to a prototype) of each pixel in the image. These vectors may be used to de-warp each exposure back to a prototype, resulting in a sequnce of images with a unifrom PSF. These images can then be globally deconvolved by an iterative process to produce a sharper result.

The side-effect of the generation of the shift maps, is that these contain information about the motion of the atmosphere. Running the shift maps generated from each exposure as a video sequence gives an impression of "optical flow" of the distorting medium.

We have applied these techniques to astronomical imagery and to horizontal or surveillance images, and adapted it to stereo disparity work. Movie sequences are provide there for your perusal.

Please contact Andrew Lambert (a-lambert@adfa.edu.au) for problems with this server.