An investigation of methods for determining depth from focus

TL;DR: A general, matrix-based method using regularization is presented, which eliminates some fundamental problems with inverse filtering: inaccuracies in finding the frequency domain representation, windowing effects, and border effects.

Abstract: The concept of depth from focus involves calculating distances to points in an observed scene by modeling the effect that the camera's focal parameters have on images acquired with a small depth of field. This technique is passive and requires only a single camera. The most difficult segment of calculating depth from focus is deconvolving the defocus operator from the scene and modeling it. Most current methods for determining the defocus operator employ inverse filtering. The authors reveal some fundamental problems with inverse filtering: inaccuracies in finding the frequency domain representation, windowing effects, and border effects. A general, matrix-based method using regularization is presented, which eliminates these problems. The new method is confirmed experimentally, with the results showing an RMS error of 1.3%. >