This is quite a complex problem, but I think you can solve it if you study the subject properly and put some decent effort. I hope you goals really worth the effort, and, additionally, you could potentially help many other people. So, thank you very much for your interest in this work and your dedication which you will certainly need.
The problem is quite solvable, because there is much more complex problem: creation of panoramic images out of a set of separate images, not using information on exact camera positions/angles. People found quite nice solution to this complex problem. Your problem can be considerably simpler, because you can consider images "flat" and probably base your techniques on the non-distorted (or distorted just a bit) images. You need to find correlations between neighboring images using their common features.
Let's start with something, which will be much closer to the
rocket surgery. From what I know so far, such problems are solved using RANSAC algorithms. Please see:
http://en.wikipedia.org/wiki/RANSAC[
^].
With C++, you can take a benefit of using the Open Source Computer Vision Library, Open CV:
http://en.wikipedia.org/wiki/Open_CV[
^],
http://opencv.org/[
^].
This is the relevant code sample I was able to find so far:
http://docs.opencv.org/doc/tutorials/features2d/feature_homography/feature_homography.html[
^].
Starting from this page, locate related library classes in Open CV documentation and learn using then. It will give you recognition of the matching features of different images. From this information, you should be able to derive relative orientation and shift between different images and ultimately stitch them together.
—SA