A hybrid parallel projection approach to object-based image restoration

Approaches analyzing local characteristics of an image prevail in image restoration. However, they are less effective in cases of restoring images degraded by large size point spread functions (PSFs) and heavy noise. The recently proposed learning based approaches perform well on recovering details from images degraded by large size PSFs, yet involves complicated implementation process and high computational expense. In this paper, we propose a hybrid approach to object-based image restoration. This method incorporates common characteristics of images from a class of objects into image restoration. These characteristics are represented as deterministic sets built on principal component analysis (PCA) models. The sets are combined with the observation model represented via a Bayesian approach to constrain the solution. A parallel projection algorithm is also proposed to find the solution that satisfies all constraints. Experiments performed on frontal face images using the proposed approach show superior performance over those based on local analysis in the cases involving large size PSF and heavy noise degradation. Compared with learning based approaches, the proposed approach can be implemented with ease and the solution can be found with less complexity.