Occluded Object Reconstruction with Partial Appearance

Objects in a scene are often occluded each other. Reconstructing their appearance from their visible parts plays an important role in object detection, image analysis, scene understanding, and depth estimation. This paper introduces an integrated approach that utilizes inpainting and image segmentation for reconstructing the appearance of occluded objects. In the first stage of the system, inpainting or filling-in the removing region is done by the patch-based method of replicating the remaining color information of an image. Localized region-based Active Contour Model (LRACM), segmentation method is applied to segment the visible parts of the occluded object from the additional image frame in the second stage. Finally, the extracted visible parts are composited onto the inpainted background scene obtained from the first stage. We demonstrate the proposed algorithm with a number of still images. Experimental results show that the reconstructed scenes are natural and plausible like real view.