As a classical inverse problem, image denoising requires some prior knowledge on the original image. Besides data fidelity, the latter corresponds to regularization constraints on the solution. So-called dual formulations can be used to solve such inverse problems. Their interest is to cast the original formulation onto an alternative optimization problem; however, they usually correspond to relatively complex algorithms. In this project, we thus want to pioneer dual-problem solving using simpler methods. Applications on biomedical images will be considered. If time permits, extensions to inpainting problems as well as multilevel iterative methods may be investigated. Requisites : courses in signal/image processing, interest for algorithmic methods and general knowledge in programming (MATLAB and/or Java).