Depth Recovery with a Block Shift-Variant Blur Model

Abstract

In the methods discussed in the literature on depth from defocus (DFD), the depth is usually assumed to be constant over a large local region, and the blur is estimated by considering the defocusing process to be shift-invariant over that local region. However, each local region is analyzed independently to estimate the depth corresponding to that local region. But splitting an image into independent subimages introduces an error in the estimate of blur due to improper selection of boundary [Sub88]. The effect is particularly pronounced when the subimage is of a small size. An image region cannot be analyzed in isolation because, due to the space-variant nature of the blurring process, the intensity at the border of a region is affected by the intensity immediately outside the region. This is also called the image overlap problem [Sub88] because the intensity distribution produced by adjacent patches of visible surfaces in the scene overlap on the image detector. The effect on the estimate of blur due to the image overlap problem can be reduced by choosing a larger size for the subimage. But it may not be always possible to choose a large enough local region owing to the depth variations in the scene.