Rotation-Invariant Correlation Features for 2d Sparse Unknown View Tomography

Shuai Huang, Deqiang Qiu

Abstract: 2D sparse unknown view tomography aims to reconstruct a set of point sources from its random projections with unknown viewing directions. We propose a set of rich rotation-invariant correlation features that can be extracted from the projections, which allows us to reconstruct Gaussian point-source signals without estimating the viewing directions. We show that the proposed correlation feature is a quadratic function of the signal, and formulate the reconstruction problem as solving a system of quadratic equations with linear and nonnegative constraints. Simulation experiments show that the proposed approach is robust to noise and successfully recovers the point set with Gaussian sources.

External IDs:dblp:conf/allerton/HuangQ23