Results from rectifying two images. The first is a tile pattern. The second is a billboard which was moved to view straight on. Not sure why the results come out distorted. I tried using Interp2 for fast results. For the mosaic, i used the slower for loops, but the results were much better.
| |
| |
Here is the result of building some mosaics. Most of the pictures were taken from my trip to San Diego for a conference. I actually have a ton more. ill post them later
 |  |  |
 |
 |  |  |
 |  | |
 |
As you can see, the last image was effected by changes in lighting
 |  |  |
 |
In the results above the images were blended. This was done by taking the bondary points and creating a gradient mask at the edges. For images that were really distorted this would be a problem, but in most cases this worked fairly well. Here is a result of the first mosaic without blending
Here are the results of the Harris interest point finder. So many!
here are the results after the non maximal suppression. At the top, is the harris points with the chosen points by nonmaximal suppression circled in blue. Below are the points by themselves.
Here is an example of the feature descriptions used to match the interest points. On the left the 8x8 descriptor. On the right, the point where the feature was extracted.
Here is an example of a result produced by selecting the matching points
Here are the results from running the Ransac algorithm to produce the homograpy matrix. I know its not pretty :( . Not sure what went wrong ... currently investigating. Since it didnt produce good results, I only did this one result
