This allows to limit the amount of neighbor information spread in the global optimization. A value of 0 will disable the decay of neighbor cost information and allow unlimited propagation. This will result in very far extrapolation of disparity values into low-contrast areas where measurements are not discriminative anymore. A value between 1 and 7 will only keep the upper n-bits when propagating cost information from one pixel to its neighbor. Thus 1 is the most aggressive reduction of extrapolation, yielding more holes, but reducing false disparity values around object boundaries to low contrast areas.
An integer (from 0 to 7) with the function described above.