Retrieving the parameters of cryo Electron Microscopy dataset in the heterogeneous ab-initio case

A cryo Electron Microscopy dataset is composed on tomographic projections of an object (e.g. a macromolecule). The projection orientation information is unknown. The scope of this paper is the projection parameterization in the case of a deformable object. An overview of the parametrization methods is presented. Then a new approach based on manifold learning is detailed. Finally, an evaluation method for each substep of the parameterization is proposed. The resulting evaluation of the different methods on a 2D synthetic database shows the efficiency of our approach.