Constrained Texture Restoration

Abstract

A method is proposed for filling in missing areas of degraded images through explicit structure reconstruction, followed by texture synthesis. The structure being reconstructed represents meaningful edges from the image, which are traced inside the artefact. The structure reconstruction step relies on different properties of the edges touching the artefact and of the areas between them, in order to sketch the missing edges within the artefact area. The texture synthesis step is based on Markov random fields and is constrained by the traced edges in order to preserve both the shape and the appearance of the various regions in the image. The novelty of our contribution concerns constraining the texture synthesis, which proves to give results superior to the original texture synthesis alone, or to the smoothness-preserving structure-based restoration

Image completion by structure reconstruction and texture synthesis

“…From that point of view, this is similar to the effect changing one pixel at coarse scales has on many pixels at finer scales. This contributes to creating boundaries that match the examples and are not only constrained by their extremities, as in [22,33]. Many publications mathematically express image completion as an optimisation problem and try to find the global optimum in an efficient manner.…”

“…One is that these methods need a set of specially tuned parameters that will work with a set of textures but not with others. Moreover, for images presenting different types of texture, it seems that the use of a structural constraint to make sure only the right texture is synthesised in the right place is primordial (Figure 1(f) and [30,25,22,33]). Another problem is that boundaries between the various synthesised textures propagated from the surroundings of the hole must look realistic in the hole.…”

“…Only simple topologies can be tackled (the provided results are all topologically simple and the algorithm cannot cope with intersecting curves). In [33], arcs of circles are used to link region boundaries and straight lines for boundaries that cannot be paired. Boundaries are paired based on circle fitting, corresponding region matching and sequentiality (to reduce the number of boundary crossings in the hole).…”

“…The segmentation of the image is an important part in identifying the different regions. In this work, we have used the segmentation method in [12] called JSEG, which we found to work well for a wide variety of colour-textured images; this method has been reported to be one of the best available segmentation methods [22,33].…”

“…The third category of methods combines the two approaches by creating texture that is constrained by structure. These methods have been successful in completing holes in images, by propagating structure into the hole and then creating texture based on that structural constraint [17,25,5,22,10,33,38,36,3,34,6].…”