Data-driven Multi-channel Super-resolution with Application to Video Sequences

A method is proposed for super-resolving multichannel data with applications to PREDATOR video sequences. Using a generalization of Papoulis sampling theorem, a closed form solution has been obtained leading to a high speed algorithm which can be realistically applied to large data sets such as video sequences. In existing multi-frame methods it is a common practice to assume that the channel transfer functions are known and invariant from one frame to another, using empirical models such as Gaussian, sinc, etc. We have assumed that the transfer functions are unknown and may vary even when the same sensor is employed, and hence use the observed data to derive the Point Spread Function PSF for each frame. The estimated PSFs are used in the super-resolution algorithm. Results on PREDATOR video images are then given.