A fast subspace method for image deblurring

Yanfei Wang, Shiqian Ma

Abstract: Image deblurring problems appear frequently in astronomical image analysis. For image deblurring problems, it is reasonable to add a non-negativity constraint because of the physical meaning of the image. Previous research works are mainly full-space methods, i.e., solving a regularized optimization problem in a full space. To solve the problem more efficiently, we propose a subspace method. We first formulate the problem from full space to subspace and then use an interior-point trust-region method to solve it. The numerical experiments show that this method is suitable for ill-posed image deblurring problems.