Collaborating Markov point processes and neural networks: Application to fresco reconstruction – ARTEAK

This project aims at significantly improving capabilities of computer aided reconstruction of very large frescoes by making collaborate existing approaches in an original way under realistic conditions. Image-based reconstruction consists in determining the optimal spatial organization between the parts of an object of interest characterized by structured visual information. Such a general problem has straightforward applications in cultural heritage restoration and archaeology but also in forensics and biology. The cornerstone of this project is to combine the power of Machine Learning (ML) approaches to discover optimal data representations with the flexibility of variational formulations to represent priors on researched solutions. Beyond the applicative context of this project, this project aims to draw multiple and original interactions between ML and optimization, either by involving ML in the computation of some terms of the functional or by involving ML in the optimization process itself.