Here is a very good link:
Matthew Brown and David G. Lowe, “ Automatic Stitching of Panoramic Images Using Invariant Functions ,” Computer Vision International Journal 74, 1 (2007), pp. 59-73.
The process is basically this:
- Extract image features at duplicate key points.
- Comparison of images with images to determine areas of overlap between images.
- Ligament adjustment to align all relevant images.
- Rendering tricks to make the result look good.
The link above uses the SIFT functions described in this article:
David G. Low, " Distinctive Image Functions from Non-Invariant Key Points ," International Journal of Computer Vision, 60, 2 (2004), pp. 91-110.
Prerequisites for understanding this material include:
In his answer, Coan offers an excellent tutorial.
It is possible to use a large amount of existing code to extract functions ( VlFeat provides the Matlab and C ++ libraries), RANSAC and minimization (I do not know what the best libraries are for them). OpenCV is also a very good computer vision library.
How much time do you have for this project? I think this is ambition. Based on my lack of computer vision experience, I think that a good goal would be to find the transformation between the two images and stitch them together. You would learn a lot from this.
user334856
source share