Re: [OctDev] New imrotate

[Stefan v. d. W. <st...@su...>]

Hi Justin

The script works well.  If you want it to be incorporated into
octave-forge

- add a license
- fix imgrotate()

octave:1> imgrotate
error: `theta' undefined near line 31 column 11
error: evaluating binary operator `*' near line 31, column 17
error: evaluating binary operator `/' near line 31, column 21
error: evaluating assignment expression near line 31, column 9
error: called from `imgrotate' in file `/tmp/imgrotate.m'

- attach it to an email (do not include it in the text body)

Regards
Stefan

On Mon, Oct 18, 2004 at 03:55:03PM +0200, Justus H. Piater wrote:
> Hi,
> 
> I've had some problems with octave-forge's imrotate.m (lack of
> precision, global gray-level alteration, inability to recover the
> rotation mapping), so I wrote my own from scratch. 
> 
> Highlights:
> 
> - 100% precise (unless you find bugs...), as it works by applying a
>   rotation homography to each pixel coordinate. The homography is
>   returned to the caller if desired.
> 
> - "nearest"-neighbor mapping or "bilinear" interpolation
> 
> - "loose" or "crop" bounding box
> 
> - faster than the existing imrotate (on the few informal tests I ran)
> 
> - does not rely on external functions not in core octave
> 
> - code is relatively compact and intuitive (to me at least...)
> 
> Caveats:
> 
> - Only a single-band image matrix is accepted as input for now.
> 
> - Its memory usage can still be improved. It makes heavy use of
>   exhaustive index lists (to take advantage of vector notation); the
>   current implementation keeps more of them around simultaneously than
>   necessary.
> 
> The function is currently called imgrotate to allow it to coexist with
> the existing imrotate. However, unless bugs show up, I don't see why
> it should not eventually take its place.
> 
> You feedback is more than welcome.
> 
> Enjoy,
> Justus
>