[GRASS-dev] New georectifying module in TclTk

[Glynn Clements]

Michael Barton wrote:

> For RMS error calculation, I just need to use either the forward or reverse,
> right? I don't see where I need both directions.

You probably don't really need both, at least for an affine
transformation. The diagonal error computed in one direction will be
proportional to that computed in the other direction. The same doesn't
apply for the x and y error, though.

For polynomial transformations, the situation is more complex. The
inverse of a polynomial transformation isn't itself a polynomial
transformation. Or, to put it another way, the best-fit reverse
transformation won't be the inverse of the best-fit forward
transformation.

At the very least, you need to make sure to use the correct direction
(rasters are usually transformed by projecting coordinates from the
target location into the image's X/Y coordinate system; at least,
that's how r.proj works).

It might help to compute the error both ways to provide an idea of how
good a fit the transformation is. A transformation which accurately
maps a minimal set of control points won't necessarily accurately map
the entire region.

-- 
Glynn Clements <glynn at gclements.plus.com>