Variational image inpainting by combination of features



Image inpainting is the process of reconstructing parts of an image in a visually plausible way. A category of methods well established in the literature is based on partial differential equations (PDE). The PDE approach consists in iteratively propagating geometric and intensity information into the missing area. However designing such models requires deep insights and mathematical understanding of the problem at hand. Based on Risheng Liu's paper for building the PDE in a bottom up manner using linear combinations of invariants, we propose an optimized implementation that we compare to previous methods.