Probabilistic multiscale image models often yield very powerful image processing and analysis algorithms requiring minimal computational resources. However, in many (perhaps most) applications of interest, the object we wish to recover is subjected to a transformation or distortion during data acquisition (e.g., tomographic projections, atmospheric distortions, blurring). In such cases, image recovery requires the solution of an inverse problem. Multiscale image models have proved to be tremendously useful in image denoising, segmentation, and texture synthesis, but less is understood about their utility in solving inverse problems in imaging. The goal of this work is to propose a unified multiscale framework for image reconstruction and restoration problems involving transformed or distorted data, treating both Gaussian and Poisson observation models.
Copyright. All Rights Reserved.