Single Image Reflection Removal

Abstract

This paper presents a proposed algorithm for removing undesired reflections from a single mixture image captured in front of transparent glass mediums. The proposed algorithm separates the single input image into background layer (actual scene) and the reflection layer by clustering mixture image pixels, using the prior knowledge that the background edges are of larger magnitude than reflection edges and the model of dichromatic reflection. Experimental results on real-world images proved that the proposed algorithm gets a result of the background image with reflections removed as much as possible. A quantitative and visual quality comparison between the proposed algorithm and state of the art algorithms is performed.