An iterative algorithm for image inpainting using aggregation functions
Manuel González-Hidalgo, Sebastià Massanet, Arnau Mir, Daniel Ruiz-Aguilera · 2017
Image inpainting is a prolific line of research due to its applications in restoration of missing or damaged areas of the image. In this paper, a novel iterative algorithm for image inpainting based on aggregation functions and penalty-based functions is presented. The algorithm combines diffusion-based and patch-based techniques. In each iteration of the algorithm, the level of consensus among the known pixels in a neighbourhood of each pixel is computed. If a minimum value of consensus is reached, the pixel is recovered by means of an aggregation of the known pixels of the neighbourhood through aggregation functions and penalty-based functions. Otherwise, a non-local search of similar patches is performed and then a similar aggregation but now of the centre pixels of those patches more similar to the region we must recover is carried out. Experiments on synthetic and natural images show the potential of this algorithm both from the qualitative and the quantitative points of view in comparison to other classical inpainting algorithms.