The use of context in image restoration

This paper approaches the problem of designing dynamic constraints for use with an ill-conditioned image restoration process. A rule-based system is proposed that exploits feedback between image analysis and the low level analytic process. This allows the system to update the constraints by reasoning about their validity in the context of an image description generated from an a-priori class model and partial restorations of the image. The system has been used to show the effect of using structural constraints to define the solution space as an alternative to static, statistically-derived a-priori bounds on the solution.