Data-driven multichannel superresolution with application to video sequences

Publisher Website
Google Scholar
Abstract
A method is proposed for superresolving multichannel data with application to video sequences. Based on a generalization of Papoulis’ sampling theorem, nonuniform samples of multiple channels are merged to generate high-resolution data. To overcome sampling ill posedness in the presence of noise, image frames are projected from standard orthonormal bases onto optimal Riesz bases defined by channel point spread functions (PSF’s). The method is therefore designed to perform under practical conditions of noise and other degradations. Unlike existing methods, where empirical models such as Gaussian, sinc, etc., are commonly used for characterizing channel PSF’s, the PSF’s are assumed unknown and possibly different and hence are blindly estimated from the observed data. The estimated PSF’s are then used to construct biorthogonal projection filters for the superresolution algorithm. This approach gives rise to a closed-form solution leading to a high-speed algorithm. The method has been tested and verified on PREDATOR video sequences (PREDATOR data are airborne video sequences of the Defense Advanced Research Projects Agency obtained by unmanned aircrafts)

This publication has 17 references indexed in Scilit: