Let’s return to the image matching problem mentioned in Chap. 6 . How can we match sub-images 7.1a and b? Can we use the corners introduced in Chap. 6 for matching? Sub-images 7.1a and b have both rotation and scale transformations, and corners are not invariant to scale. Therefore, we need to find a new type of feature point (keypoint) for matching images under complex transformations. If we can locate some regions of interest in the image centered on this feature point, such as the windows shown in 7.1c and d, and if we know that these two regions match, we can calculate the rotation and scale transformations between them, thereby matching the two images. However, not every region in the image can be used for matching as a feature point area.