Continous Image PRocESSIng: models and algorithms – CIPRESSI

One of the great challenges of imaging sciences is to recover high resolution images from incomplete and possibly noisy measurements. A common practice when tackling such inverse problems is to define a pixel grid on which to reconstruct an image that accounts for the observations, for instance by minimizing an energy. There are many drawbacks to that approach: discretization artifacts, large problem sizes, difficulties to analyze the properties of the model...
With CIPRESSI, we propose to develop novel discretization methods which respect the continuous nature of imaging problems. By leveraging the properties of continuous models, we will design fast algorithms which produce artifact-free images.
Besides laying the foundation of this next generation of imaging methods, we will apply them to two real problems, in meteorology and microscopy.