Generation of depth image through interpolation and extrapolation of intermediate images derived from stereo image pair using disparity vector fields

摘要

A depth image is generated from a pair of stereo images by means of a disparity vector field-based digital image processing methodology. Specifically, images corresponding to missing angular views are derived from the stereo image pair based upon an iteratively refined set of disparity vector fields that associate desired missing image locations with the pair of stereo images. For a reduced angular separation between the optically recorded pair of stereo images, the stereo images are employed as the images corresponding to intermediate angular viewing locations, and images corresponding to missing angular views are reconstructed by either digitally interpolating or digitally extrapolating the stereo image pair, depending on whether or not the missing angular view is between the angular views of the stereo image pair or outside the angular views of the stereo image pair. The missing images are then interlaced with the images of the stereo to derive the depth image. A system of nonlinear equations is formed relative to an unknown current estimate of the disparity vector field corresponding to the missing image. This system of nonlinear equations is then linearized from which initial estimates of the disparity vector fields are derived relative to the current estimates of the disparity vector fields. Solutions to the resulting set of linear equations are used to iteratively refine estimates of the disparity vector fields.