Angular difference measure between tomographic projections taken at unknown directions in 2D

This paper introduces a new measure for estimating the angular difference between two tomographic projections belonging to a set of projections taken at unknown directions. The measure is potential for many applications such as projection refinement or projection classification, which are important in the process of tomographic reconstruction. Our measure relies on the construction of a neighborhood graph for projection moments, the calculus of the angular difference for neighboring projections and the computation of geodesics on this graph. The accuracy and the robustness of our measure is shown on a test database including 50 2D gray-level images at different resolutions and with different levels of noise.