For the pinhole camera model, the mapping from 3D to 2D coordinates is described by a perspective projection(rectilinear projection). However for the image stitching application, perspective projection will bring some problems. Its primary disadvantage is that it can greatly exaggerate perspective as the angle of view increases, leading to objects appearing skewed at the edges of the frame. The input images: enter image description here

Stitched image: enter image description here

The images are from the link: https://www.cambridgeincolour.com/tutorials/image-projections.htm

I can't figure out the reason that the left port almost doesn't change much but the right port undergoes much change for the the third image. The overlapping part of the second and third images doesn't change a lot.


I draw an diagram to show capturation relationship between two images. enter image description here

The scene contains two black lines. The red square indicates the first capturation position and the blue square shows the second capturation position. To match two images, the second image has been warped as the following figure shows. It looks like the same transformation that the hall example undergoes. I don't understand why such transformation is required to match the two images. enter image description here


closed as off-topic by Kyle Kanos, Jon Custer, glS, Anders Sandberg, sammy gerbil Jul 25 '18 at 18:07

This question appears to be off-topic. The users who voted to close gave this specific reason:

  • "This question appears to be about engineering, which is the application of scientific knowledge to construct a solution to solve a specific problem. As such, it is off topic for this site, which deals with the science, whether theoretical or experimental, of how the natural world works. For more information, see this meta post." – Jon Custer, glS, Anders Sandberg
If this question can be reworded to fit the rules in the help center, please edit the question.

  • $\begingroup$ I'm voting to close this question as off-topic because it is asking about image processing, not physics. $\endgroup$ – sammy gerbil Jul 25 '18 at 18:07

The rectilinear projection keeps all straight lines straight. As a result, the image stretches increasingly stronger toward the far edge. The answer to your question is that the right side of the right image is much closer to the right edge of the stitched image and for this reason is stretched more.

In contrast, some other projections avoid this effect, but anavoidably at the expense of at least some straight lines becoming curved. Examples include the obvious fish-eye projection and the cylindrical projection shown on the same site. Note that, while the image is not nearly as much stretched on the sides, the roof line is curved, unlike in the rectilinear projection in your question:


These are just natural creative trade-offs of geometric optics.

  • $\begingroup$ Thanks for your reply. The question has been edited to provide more information. $\endgroup$ – Jogging Song Jul 21 '18 at 1:58
  • $\begingroup$ This transformation is required, because these two images were taken from two different angles (the camera was rotated between the shots). When you rotate the camera, the perspective changes. It changes in such a way that straight lines remain straight, but angles between them change. So, to stitch these two images, you first must apply the transformation that compensates for the change in the camera angle. $\endgroup$ – safesphere Jul 21 '18 at 3:20

Not the answer you're looking for? Browse other questions tagged or ask your own question.