Reflection removal based on GMM

Image reflection is often a problem in the process of taking pictures through glass windows, how to separate the objects on both sides of the glass is an important but ill-posed problem. In this paper, we study several algorithms of removing reflection. Focuses on the ghosting formation model which assuming the glass is two layers or has a certain thickness(3-10mm), the object refraction and reflection several times through the glass will form a spatial offset. The layer separation algorithm regularized with a Gaussian mixture model. We use three methods to test the synthetic images and real-world images with different reflective effect. The advantages and drawbacks of each algorithm are outlined and evaluated.