The algorithm searches for corresponding image patches in the supplied
stereo pictures by computing the correlation between small image-patches of
size 7x3 pixel. The calculated disparity values assume only integer values
Fig 1: Output of the first stage of the stereo-algorithm
In a second processing step, these integer values can be refined to
sub-pixel precision by using an iterative scheme proposed by Okutomi and
Kanade. (further reference: Okutomi & Kanade, Int. J. of Comp. Vis. 7:2, 143-162,
1992 ). A fixed window-size is used,
so this second interpolation step is essentially identical to a subpixel-shift
detection as it is used in standard optical flow methods.
Fig 2: Refined Disparity Map
© 1994-2000 - all rights reserved.