A Statistical model for the warp matrix in super-resolution video reconstruction

This paper advances in the analysis of the least mean square (LMS) algorithm applied to super-resolution reconstruction (SRR) of an image sequence. SRR of image sequences is highly dependent on the quality of the registration stage. When motion between frames is unknown and has to be estimated, the best available statistical model still requires the numerical estimation of the moments of the registration (warp) matrix. In this work we derive an analytical model for these moments for Gaussian registration errors. The new model allows for different boundary conditions. Monte Carlo simulation results illustrate the accuracy of the new analytical model.