Revision history [back]

The ROI you have is (per images) the region where all pixels have been correctly rectified. IE, the ROI is made by:

• top left corner: first pixel where a full row and a full colon as been remapped (no black pixel),
• top right corner: the Y is the same as the previous, the X coordinate is the latest pixel where a full colon can be used,
• bottom right corner: the X is the same as the previous, the Y is the latest pixel where a full row could be used,
• bottom left corner: The Y is the same as the previous, the X is the same as the top left corner.

This is why ROI are not the same on left and right images, and you have to specify them to the Stereo algorithm (at least Block Matching) and not directly apply them to the images before stereo reconstruction (images won't be aligned anymore if you do that and an error will be thrown as the images won't have the same size anymore...)

This is my understanding of the ROI, but for a formal definition, you should probably say that this is the biggest included (aligned to the axes) rectangle containing only valid pixels after stereo rectification.