Remove Reflections using Bisquare Iterative Reweighted Least Square
Reflections are ubiquitous effects in photos taken through transparent glass mediums, and represent a big problem in photography that impacts severely the performance of computer vision algorithms. Reflection removal is widely needed in daily lives with the prevalence of camera-equipped smart phones, and it is important, but it is a hard problem. This paper addresses the problem of reflection separation from two images taken from different viewpoints in front of a transparent glass medium, and proposes algorithm that exploits the natural image prior (gradient sparsity prior), and robust regression method to remove reflections. The proposed algorithm is tested on real world images, and the quantitative and visual quality comparisons were proved the better performance of the proposed algorithm on an average of 0.3% improvement on the blind referenceless image spatial quality (brisque) error metric than state of art algorithm.