Stereo image rectification projects images acquired from the left and right cameras so that the images reside in the same plane. The rows of rectified images align perfectly so that a point falls on the same row in both in the left and the right images.

By row-aligning matches, stereo image rectification simplifies the processes of calculating stereo correspondence and computing a disparity map. Stereo image rectification is based upon the spatial relationship between the cameras, which is obtained from information produced during stereo calibration.

To speed up the rectification process, a look-up table can be computed for each camera to interpolate points from the original image and create a new rectified image. Because learning a look-up table is memory intensive, the process is optional in Vision.

Vision also accepts a scale parameter. Valid values are 0-1. A value of 1 constrains the rectified images to the original image size. Values smaller than 1 increase the scale of the rectified images.

The following figure illustrates the process of image rectification: