Restoration of multiple misregistered images

Abstract

Summary form only given. A stochastic modeling approach to the restoration of misregistered corrupted images of the same scene has been adopted. If a wide-sense stationarity assumption is made on the 2-D continuous version of the image of the scene, it can be seen that individual discrete image samples from this continuous image exhibit similar statistical properties in the spatial domain, i.e. the (auto) covariance matrix defined on each discrete image is identical. The temporal correlation between a pair of discrete images is dictated by the temporal displacement between images and the stochastic model in the continuous domain. This temporal correlation can be expressed in terms of the misregistration of a pair of discrete images and the correlation matrix of the discrete image. Hence, the stochastic model-based approach defines a parametric form of the spatio-temporal correlation matrix, in terms of the spatial model parameters over a single image lattice and the displacement values between different frames.<>