Image Restoration with Morphological Erosion and Exemplar-Based Texture Synthesis

Image restoration provides a means to reconstruct damaged regions of an image. The paper presents a new restoration algorithm, which implements the filling of damaged region with morphological erosion, and propagates structure/texture features of the known region into the damaged region with exemplar-based texture synthesis. The method can retain the continuity of image isophotes between the known region and the restored damaged region, and output a complete, natural-looking image. Through comparative experiments with some existing methods, we demonstrate the effectiveness of the algorithm in removing large objects as well as thin scratches.