LSR based astronomical image denoising via adaptive dictionary learning

Motivated by local coordinate coding (LCC) theory in nonlinear manifold learning, we proposed a new image representation model called local sparse representation (LSR) for astronomical image denoising. Meanwhile, a fast approximated LSR method by first performing a K-nearest-neighbor search and then solving a l 1 optimization problem is presented under the guarantee of denoising performance. In addition, we incorporate the LSR model and adaptive dictionary learning into a unified optimization framework, explicitly establish the inner connection of them. Such processing allows us to simultaneously update sparse coding vectors and the dictionary by alternating-optimization method. Our experimental results have shown convincing improvements on astronomical image denoising.