| Pair 1 | ||||
|---|---|---|---|---|
| Pair 2 |
For part II, I build an image stitcher given a working corresponder. I use a wetware corresponder.
I'm also given a least squares program by Prof Berg.
The initial task of entering correspondance points on the source image comes from the mouse point entering method in the given sample code. Similarly for the target image.
The sample code comes with homogeneous matrix multiplication and division methods to calculate the tranformation matrix.
As Prof. Berg demonstrated in class, with minor mathematical manipulation you can go from scalar-scalar maps to vector-vector maps.
Transforming an entire image myself with a matrix would not have been fun. Prof Berg suggested using MATLAB libraries. A quick google search and http://www.mathworks.com/help/toolbox/images/ref/imtransform.html tells me how to do it.
A small snag comes in that MATLAB is too smart and centers transformed images destroying translation. I googled the fix: http://blogs.mathworks.com/steve/2006/07/07/spatial-transformations-translation-confusion/ .
Finally, average the images. Rather than do it myself, I get alpha to do it as seen here: http://www.mathworks.info/matlabcentral/newsreader/view_thread/297276 .
As can be seen by the pics, the fusion is aligned.