Separating Reflections from Polarized Images using Alpha Matte Method includes Image Smoothening

Abstract

When we take a picture through transparent glass, the image obtained is often a linear superposition of two images: The image of the scene beyond the glass plus the image of the scene reflected by the glass. In this paper, our ultimate aim is to separate the effect of reflection from such images captured behind glass. To accomplish our goal, we are going to take multiple polarized reflected images which are taken at different polarizing angles with same transparent medium such as glasses as input and aim to separate the reflections from those images to acquire the original background layer without reflections and the original reflection layer. Here we incorporated a technique called alpha matte method to separate the reflections. We also consider our work as an optimization problem to exploit the mutually exclusive information from our input polarized images to achieve high quality reflection separation results. We test our approach on various images and found that our approach can generate good reflection separation results than the existing methods without any prior knowledge about the images. We also embedded an image smoothening process before final output image generation to achieve high quality noise-free output.