Stereo matching framework
1. For every disparity, compute raw matching
costs
Why use a robust function?
occlusions, other outliers
Can also use alternative match criteria