Astronomical images taken by ground-based telescopes suffer degradation due to atmospheric turbulence. This degradation can be tackled by costly hardware-based approaches such as adaptive optics, or by sophisticated software-based methods such as lucky imaging, speckle imaging, or multiframe deblurring. The latter involves processing an entire sequence of observed images of an underlying celestial body to recover a single sharp image. This recovery is complicated by two major difficulties: (i) the number of observed frames can be very large; and (ii) the blur corrupting each of the observed frames is unknown.
We tackle both these difficulties by developing a new algorithm that incrementally processes the observed frames to obtain a deblurred, high-quality image. Although the associated optimization task is non-convex, the incremental processing reduces to solving simpler, convex subproblems. Of course, these subproblems must be rapidly solved to obtain an overall efficient method; so we also discuss how to solve the subproblems. Putting together all details we obtain a simple, but efficient and versatile multiframe deblurring algorithm.
Encouraging results on simulated and real-world images demonstrate that our method yields deblurred images of comparable and often better quality than existing approaches.
based on joint work with S. Harmeling, M. Hirsch, D. Kim, B. Schölkopf
Back to Workshop V: Applications of Optimization in Science and Engineering