Two-step Algorithm for Image Inpainting.

In this work we propose a new algorithm for image inpainting. The proposal is patch-based, so we look for similar small regions (windows) through the whole image to inpaint the unknown area. The final goal is to obtain a complete image with no visual differences between the original part and the reconstructed one. The main novelty is the use of color and gradient properties to look for similar windows of the image. We combine these two properties in two stages. In the first one we cluster all the available windows taking into account only the gradient feature. With this step we preselect some windows from the set of all available ones. From all the preselected ones, we finally select the most similar in color intensity. The results show that our algorithm gets final images with better textures than the ones obtained just considering color features.