We have developed a fast stereo matching method using fast correlation
and dynamic programming techniques in the coarse-to-fine framework.
The algorithm produces a reliable dense disparity map. The fast
cross-correlation was realized by using the box-filtering
technique. The time spent in the stage for obtaining the mean and
standard deviation for the normalized cross-correlation is almost
invariant to the search window size. The typical running time for a
512 
512 image is in the order of minutes rather than hours.
By using the zero-mean normalized cross correlation (ZNCC) similarity measure rather than the simple SSD or SAD, the reliability of the algorithm was increased. The algorithm was shown to be reliable by testing on several different types of images.