Image Inpainting Based on Local Patch Search Supported by Image Segmentation

Sarah Almeida Carneiro, Hélio Pedrini, Silvio Jamil Ferzoli Guimarães

Abstract: Image inpainting can be defined as a restoration process in which damaged or selected regions are repaired by taking into account the image content. In this work, we employ a local-based strategy instead of a global one to identify the best existing patch with information to replace the damaged/selected patch. In order to properly identify the most representative patches, we propose a method based on the (i) creation of a local graph using a similarity of patches in the original image and (ii) partition of the image into regions according to hierarchical image segmentation to support the local patch identification. The experimental results demonstrate that our local search outperformed the results of image inpainting in terms of both qualitative and quantitative aspects, when compared to global search of patches.