Multiresolution analysis constitutes a general framework both for image modeling and for developing efficient image analysis algorithms. We will first discuss how multigrid numerical algorithms can effectively solve a wide range of image segmentation and image inpainting problems formulated using diffusion partial differential equations. We will then present hierarchical stochastic models for natural images which employ wavelet-domain scale-recursive Markov structures and show how they can be applied for image inapainting. We will also report on our ongoing effort to apply these techniques in practice on the problem of digital restoration of missing parts in the pre-historic wall paintings of Akrotiri-Thera.
Talk slides in pdf [~7,5MB]http://www.iit.demokritos.gr/docs/seminars/papandreou_demokritos09.pdf