image resize in image stitching pipeline

Question

OpenCV contains one image stitching module. The pipeline is followed

There is one step named Resize. As I understand, the step is added in order to reduce computation because feature detection for one big image may need a lot of time. However, the homography calculated from one pair of two smaller images can't apply to the original images. Any help is appreciated.

Answer 1

Since the homography matrix can be perceived as a transformation matrix; the homography matrix which is estimated for the scaled image can be used with original size image as well.

H_orig = H_scale * H_estimated is a valid operation. You can read more about the homography matrix and what its components stand for from here and here.