Stereo Correspondence

Assumptions

• Assume parallel (co-planar) image planes…
• Assume same focal lengths
• Assume epipolar lines are horizontal
• Assume epipolar lines are at the same y location in the image

Soft constraints

• Similarity: The corresponding pixel should have a similar intensity
• Uniqueness: There is no more than one matching pixel
• Ordering: Corresponding pixels must maintain their order: pixels ABC in the left image must likewise be matched to A'B'C' in the right image.
• Limited Disparity Gradient: The depth of the image should not change too quickly.

To find matches in the image pair, we will assume:

• Most scene points visible from both views
• Image regions fro the matches are similar in appearance.

Dense Correspondence Search

This is based on similarity constraint.

Dense here indicates that we will try to find matches for every pixel in the image.

For each pixel (or window) in the left image:

• Compare with each pixel (or window) in the right image along the epipolar line.
• Choose the position with the minimum "match cost". This can be determined by sum o fsquare differences (SSD) or the normalized correlation.

Uniqueness Constraint

The matching pixel could be zero because of occlusion. Certain pixels will only be visible from one side at occlusion boundaries.