Performance of optical flow techniques
International Journal of Computer Vision
| Hi-index | 0.00 |
A novel method for visualising atmospheric turbulence is described, based on a point-by-point registration to subpixular accuracy of multiple telescopic images of scenes distorted by the atmosphere, both astronomical and terrestrial The registration, which produces detailed x and y shift-maps over the space of each image in the sequence, is based on hierarchical, windowed area-correlation. Each x and y shift-map is then used to dewarp the corresponding raw image in the sequence. Registration is made to a prototype average image, which is initially a motion-blurred average of the uncorrected image sequence, but which is Later refined to be the average of the dewarped sequence. The process is repeated, iterativeLy, using the current best average result as a new prototype. which represents a motion-blur compensated (restored) image of the scene. The resuLting x and y shift-map sequence, viewed as a movie, provides a striking visualisation of the wavefront distortion due to atmospheric turbulence.