[ptx] Re: [PanoTools-devel] PTmender, and update

[Pablo d'Angelo]

Daniel M. German wrote:

> The algorithm in PTstitcher for seam calculation is really bad, and
> creates seems in very strange locations (it is already implemented, by
> the way. Feathering will also help "ease" the transitions.
> 
> I would prefer a seam that is computed in the following manner:
> 
> * For each photograph, choose a pixel that is closest to its center
>   (center as before the panorama transformation).
> 
> My assumption is that the "best" pixels tend to be closer to the
> center of a given photograph. Of course there is no way to avoid
> ghosting this way.

Earlier version of nona have done this. However, it also results in straight
 lines.
See: http://thread.gmane.org/gmane.comp.misc.ptx/1396/focus=1435

> 
> My main problem in Panoramas has been vignetting. Vignetting will be
> removed by doing this (I haven't tested the vignetting feature in
> Hugin yet).

For me, it has worked quite well. It can even estimate the required overlap
from the overlap.

However, the circle distance approach will not correct the vignetting, it
will merely ensure that the borders of two images will have the same amount
of vignetting. In areas where more than two images cross, this will probably
break down. Plus one can still see the banding in blue skies.

> This is a broad overview of the algorithm.
> 
> Each photograph is scanned from left to right, then from right to
> left, top to bottom and bottom to top.
> 
> The result of this scan is, each pixel, a counter keeps its minimal
> distance (in pixels) from the edge of the photo (remember this
> distance is either horizontal or vertical, not optimal).
> 
> A final scan is done (line by line) where for each position, the pixel
> with the maximal distance is chosen.
> 
> This is why the seams created by PTstitcher are usually created of
> straight lines.

This is similar to the algorithm used by enblend and nona.

ciao
  Pablo