I may be able to work out the inverse mapping equation from what
Dersch has on his page. It looks straight forward enough - just have
to convert R into X and Y components (assuming the angles in the
input and output domains are the same, as that is what a radial-like
distortion means)
The forward equation is not easily achieved, but I would think there
will not be such extreme distortions as in perspective and arc. If
there are, then one would just have to live with point or better
bilinear interpolation. However, my main issue is how one goes about
figuring out what the a,b,c,d constants should be!!!!
Fred
Wolfgang Hugemann on wrote...
| Playing arount with the "new" -distortion option, I was surprised that
| it doesn't support barrel / pincushion distortion. Wouldn't that be easy
| to involve?
|
Actually there are a huge number of distortions that have yet to be
added. Including those.
I have been concentrating on 'quality' improvements, before adding more
distortions. Though that is on hold due to the pressure of other
commitments in my life, a trip to China for the whole of April being
and the Weifang International Kite Festival, being the main one.
That however should not stop you or others contributing new distortions.
Though I recommend you read up on General Distortion Techniques
http://imagemagick.org/Usage/distorts/#summary
So as to get an understanding of what is going on.
What is needed for a distortion function...
1/ The function to convert a i,j (destination) -> x,y (source) coordinate.
See the reasons in Reverse Pixel Mapping
http://imagemagick.org/Usage/distorts/#mapping
This basically defines the distorting in programmable terms.
2/ The scale changes needed for an image at that point..
that is how much the original image is compressed (and optionally rotated)
This can be expressed as 'none' (point interpolation - not good)
or as a direct scale, or as a vector scaling
di -> dx,dy vector and dj -> dx,dy
to form the 'resampling' ellipse. See the diagram in
http://imagemagick.org/Usage/distorts/#area-resampling
Though mathematically the arrows are reversed, as we need to
determine the ellipse size, so we can then map the colors in the
direction indicated.
Without this you will see the pixel mapping problems at extreme
distortions, such as IM has with the old -implode operator
http://imagemagick.org/Usage/distorts/#implode
3/ And finally if you like the +distort to handle automatic resizing of
the resulting image, you need a method of determining how big to
make the final image from a give source image, that as a x,y offset
and a specific w,h size.
Without this, macro scale resizing of the resulting image, such as
what -rotate does, does not happen.
See the SRT distortion http://imagemagick.org/Usage/distorts/#srt
The first is required, but can be difficult. For example the 'Bilinear'
distort is currently mapped in a reverse direction as it is a difficult
operation to 'reverse' mathematically.
The other two 'needs' are optional.
However for heavy distorts, such as 'perspective' the second option is
very much a MUST. Again see
http://imagemagick.org/Usage/distorts/#area-resampling
And sometimes for a distortion that completely remaps the coordinate
system, the third need is not optional at all. For example the 'Arc'
distortion always performs the third (macro image resize) regardless of
the '+' or '-' sign of the -distort operation, otherwise it is not as
simple to use as I have designed it.
| There are astonishingly few programs on the market that perform this
| sort of transformation / correction. Recently, I have been experimenting
| with PTlens and found its results in short focal length correction
| surprising. (Or, to put it the other way round: I was surprised by the
| extend of distortion that normal digital cameras produce at short focal
| lenghts.)
|
Yes I know. I have planned it, just not implemented it.
| The good thing about PTlens is that the correction factors for most
| digital cameras are already in its database and that it readily extracts
| the focal length from the EXIF data and corrects the photo accordingly.
| (And that it allows batch processing.)
|
A database of lens distortions is not something IM would probably get
into.
| I instantly thought that this could also be done with IM (without having
| to use the -fx operator). But it can't at the moment -- right?
|
Right. But if you have a bit of programming skill, you could give it a
try, or just workout the details of the above 'needs'.
Anthony Thyssen ( System Programmer ) <A.Thyssen at griffith.edu.au>
_______________________________________________
Magick-users mailing list
[email protected]
http://studio.imagemagick.org/mailman/listinfo/magick-users
