Depth-wise image inpainting

Image inpainting in a clutter-scene containing objects at several depths could be more challenging in contrast to conventional image inpainting problems. The basic inpainting problem is filling up the missing parts of an image caused by removing undesired objects existing in the foreground layer. In this paper we propose a method to inpaint missing portions of an image in different depth layers. The method uses an object retrieval method to find the best match and fills up the missing area of the incomplete query image using photometrically and geometrically registered view of the retrieved object. We create several sets of cluttered scenes composed out of 24 objects to demonstrate the efficiency of the method.