In this assignment, we had to combine two or more images into a panoramic view. This was implemented in MATLAB using the Direct Linear Transformation (DLT), normalized DLT, and Gold Standard algorithms. For the Gold Standard algorithm, I minimized the Sampson approximation to the geometric error. I tested the algorithms using 4-point and 10-point correspondences between a pair of images. The images I used are:
| Input image to transform | Base image |
|
|
|
Below are the correspondence points I used in the 4-point case, followed by the images and errors obtained using the three different methods:
| Correspondence points (input image) | Correspondence points (base image) |
|
|
DLT: symmetric transfer error = 3.0649e-022
NDLT: symmetric transfer error = 1.7368e-026
Gold Standard: Sampson error = 4.7129e-027
Here are the correspondence points I used in the 10-point case, followed by the images and errors obtained using the three different methods:
| Correspondence points (input image) | Correspondence points (base image) |
|
|
DLT: symmetric transfer error = 6.0571
NDLT: symmetric transfer error = 5.9495
Gold Standard: Sampson error = 1.4902
The code can be downloaded here.