Image recovery using sparse reconstruction based texture refinement

We present a robust algorithm for spatial recovery of missing region in images. The algorithm consists of two stages: sparse modeling and patch based refinement. We note that a model based image recovery might not be able to reconstruct the richness or details in a signal unless the signal truly fits that model. We show that the reconstruction using a sparse model provides enough information about the inherent features present in the unknown area, using which, a patch based refinement process can replicate the structure and the natural texture from the surrounding available samples. The developed algorithm is tested on a variety of image characteristics. Significant objective and subjective gains are observed compared to the state-of-the-art.