DATE:

Friday, June 12, 2020

TIME:

3:00 PM

LOCATION:

Virtual Zoom Conference

SPEAKER:

Dr. Jerome Gilles, Department of Mathematics and Statistics, San Diego State University

ABSTRACT:

In a previous work, we suggested using the Fried kernel to perform image deconvolution of stabilized images acquired through atmospheric turbulence. Despite that this approach provides very good results, it has a major drawback: four parameters must be provided. If it is reasonable to consider that three of them can be known, the fourth one (the C_n^2 describing the turbulence intensity in the atmosphere) is most challenging one to guess. In our original work, we used some brute force approach to estimate that parameter making the method computationally very expansive and very hard to use in practical applications. In this new work, we propose to reformulate the Fried kernel in a simpler way. The new formulation is more flexible and only have two parameters which can easily be estimated via a Newton descent approach. We then use a procedure which alternates a very fast state-of-the-art deconvolution algorithm with the optimization of the new Fried kernel parameters. We show, both on simulated and real images, that our algorithm has good performance (it generally over performed all existing approaches in the case of strong turbulence) and is really fast.

HOST:

Jose Castillo

VIDEO: