The CPU implementation of the block-matching method of stereo matching allows you to specify a minimum disparity and a disparity range, which lets you only search for matches within a particular range of depths. However, the GPU implementation of block matching does not allow you to specify a minimum disparity. The minimum disparity is fixed at 0, and so to make sure you search within the depth range of interest, you have to search from the minimum depth all the way to infinite depth (or, 0 disparity up to maximum disparity). This means you have to specify an unnecessarily large disparity range to search, which slows down the algorithm and increases the chance of bad matches.
So my question is, how can I get around this? Is there some standard way to preprocess or alter the images to achieve a more restricted disparity search range?