Computing depth from out-of-focus blur using a local frequency representation

Abstract

We present a method to compute depth from the amount of defocus in two images obtained from the same view-point but with different camera parameter settings. The change in defocus (blur) between the two images is proportional to the depth in the scene. We introduce a novel method to estimate the blur using a multiresolution local frequency representation of the input image pair. A confidence measure is used to discriminate between high error and low error blur estimates. Quantitative experimental results are shown for both real and synthetic images.