Recovering the Subpixel PSF from Two Photographs at Different Distances

Abstract

In most typical digital cameras, even high-end digital single lens reflex ones (DSLR), the acquired images are sampled at rates below the Nyquist critical rate, causing aliasing effects. In this work we describe a new algorithm for the estimation of the point spread function (PSF) of a digital camera from aliased photographs, that achieves subpixel accuracy. The procedure is based on taking two parallel photographs of the same scene, from different distances leading to different geometric zooms, and then estimating the kernel blur between them.