Entropy-based depth from focus

Abstract

The two-image depth-from-focus problem is reconsidered in terms of entropy loss in a linear filter. It is shown that this formulation leads to a relatively simple solution whose variance is equal to or less than that from a regression approach. The formulation is appropriate even when the point-spread function of the optical system is not well suited to a low-order regression fit or when both images used contain some degree of defocusing with distance.