|
Image processing - task no.3
author: Petr Sladek (sladep1) , Milan Kratochvil (kratom4) at CTU Prague, FEE
The Basics:
- A matrix calculator (matrix of collineation)~spoctiA,
HTML: src/amat.html |
mfile: src/amat.m
- Derivation of inverse collineation (how to get inversed A for collineation, is it inv(A) okay?):
mfile: HTML: src/amat.m
- Collineation [x1 x2]=collin(A,[u1 u2]), u1,2 x1,x2 are matricies m x n
HTML: dat/collin.html |
mfile: src/collin.m
- eval - utility - get some transformed pixels from T matrix (image transf. matrix)
mfile: src/eval.m
Diffrerent types of the image collineation methods (demos):
- "crude" point-to-point forward collin with no interpolation
- point-to-point backward collin using inverse collineation
Core task - interpolated collineation
- Collineation using bicubic interpolation, applied on the Grafitti image
Comparison of the different passpoints locations:
- HTML: dat/demo.html |
mfile: src/demo.m
- the most important is:
- very close passpoints results in high inaccuracy
Experience: do not put passpoints very close, spread them over the whole image.
- passpoints in one straight line results in the SVD malfunction (high numeric
error)
Experience: do not put passpoints into one line (avoid any linearities)
- more than 4 passpoints spread over the whole image is good practice
Image fixing (skyscraper):
| |
|