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