A set theoretic approach to object-based image restoration

The approaches that analyze local characteristics of an image to impose constraints are prevailing in image restoration, but they are less effective in the cases of severe degradation and heavy noise corruption. In this paper, we incorporate the common characteristics that the images of a class of objects present into image restoration, termed as object-based image restoration. The characteristics are represented as deterministic sets, which is combined with the set describing the image degradation model in a set theoretic formulation. A parallel subgradient projection algorithm is applied to find the solution in the intersection of these sets. Experiments performed on frontal face images show the improved performance of the approach in both mentioned cases, by comparing with local analysis based algorithms. With the concise formulation and the efficient algorithm, the object-based restoration can be implemented with ease and resolved with less complexity.