Weak images enhancement using nonlinear filtering by sampling distributions.

The paper discusses the synthesis of nonlinear filtering methods for enhancing weak images typical for such areas as terahertz and infrared vision, optical and X-ray imaging etc. The proposed approach to the synthesis is based on the principles and methods of machine learning, understood as learning from the samples of the registered data. For these purposes, a special representation of images has been developed using controlled size samples of counts (sampling representations). Based on the specifics of this representation, a generative model of an ideal image is accepted, which is then concretized to a probabilistic parametric sampling model in the form of a mixture of components. It is shown for the proposed general generative model that the image enhancement problem can be reformulated into the problem of finding maximum likelihood estimates. By covering the image surface with a system of patches, the general mixture component model is reduced to a partition model of local density supports, which allows the synthesis of computationally realistic filtering algorithms.