Lanczos interpolation

[lechristophe]

Is there any ImageJ plugin that performs Lanczos interpolation/resampling ?
I saw that the "Interpolate" option in the "Adjust size" command uses
bilinear interpolation, which creates aliasing when I reduce sizes of
photographs (I use IJ to generate thumbnails...). There is the BIG Resize
plugin (http://bigwww.epfl.ch/algorithms/ijplugins/resize/) that uses "least
square" method, but how does it compare to Lanczos in terms of quality for
1) downsizing or 2) upsizing ?

Christophe

[Michael Schmid]

Hi Christophe,

I am not aware of any sync or sync-based interpolation plugin
(Lanczos is based on a sync*sync kernel).
What you could try is the TransformJ package
   http://www.imagescience.org/meijering/software/transformj/index.html

According to the paper on that page, it should be comparable
in quality to sync faster.
A pity that the source code of TransformJ is not available...

Michael
________________________________________________________________

On 17 Apr 2008, at 22:57, Christophe Leterrier wrote:

> Is there any ImageJ plugin that performs Lanczos interpolation/
> resampling ?
> I saw that the "Interpolate" option in the "Adjust size" command uses
> bilinear interpolation, which creates aliasing when I reduce sizes of
> photographs (I use IJ to generate thumbnails...). There is the BIG  
> Resize
> plugin (http://bigwww.epfl.ch/algorithms/ijplugins/resize/) that  
> uses "least
> square" method, but how does it compare to Lanczos in terms of  
> quality for
> 1) downsizing or 2) upsizing ?
>
> Christophe

[Burger Wilhelm]

Christophe,

I have implemented various Lancsos interpolators for ImageJ for our book
examples (in www.imagingbook.com, Sec. 16.3.6) and I can send you the
corresponding Java code if you like. However, despite its recent
popularity, Lancsos interpolation does not seem to carry any particular
advantage (quality and performance-wise). For good results I would
instead recommend the Catmull-Rom spline interpolation or the
Mitchell-Netravali approximation methods.

Wilhelm.

> -----Original Message-----
> From: ImageJ Interest Group [mailto:[hidden email]] On
> Behalf Of Christophe Leterrier
> Sent: Thursday, April 17, 2008 10:57 PM
> To: [hidden email]
> Subject: Lanczos interpolation
>
> Is there any ImageJ plugin that performs Lanczos
> interpolation/resampling ?
> I saw that the "Interpolate" option in the "Adjust size" command uses
> bilinear interpolation, which creates aliasing when I reduce sizes of
> photographs (I use IJ to generate thumbnails...). There is
> the BIG Resize
> plugin (http://bigwww.epfl.ch/algorithms/ijplugins/resize/)
> that uses "least
> square" method, but how does it compare to Lanczos in terms
> of quality for
> 1) downsizing or 2) upsizing ?
>
> Christophe
>

[pcloetens]

Hello,
We noticed the following weird behavior after using Auto from the
Brightness/Contrast Tool using a non-rectangular ROI.
The pixels inside the rectangle containing the ROI but outside the ROI
itself get modified in an unexpected way. The result is quite beautiful
but not really wanted. The problem seems to occur only if the image was
inverted just before.
To reproduce the problem:
- File/New/Image...
        create image filled with Ramp
- Edit/Invert
- make any non-rectangular selection
- press the 'Auto' button in the Brightness/Contrast tool
The problem does not occur with
        Process/Enhance Contrast

Peter

[Gabriel Landini]

On Wednesday 07 May 2008, Peter Cloetens wrote:

> Hello,
> We noticed the following weird behavior after using Auto from the
> Brightness/Contrast Tool using a non-rectangular ROI.
> The pixels inside the rectangle containing the ROI but outside the ROI
> itself get modified in an unexpected way. The result is quite beautiful
> but not really wanted. The problem seems to occur only if the image was
> inverted just before.
> To reproduce the problem:
> - File/New/Image...
> create image filled with Ramp
> - Edit/Invert
> - make any non-rectangular selection
> - press the 'Auto' button in the Brightness/Contrast tool
> The problem does not occur with
> Process/Enhance Contrast

Yes, I can confirm that in 1.41c with the polygon or the freehand selections.

G.

[Wayne Rasband]

> Hello,
> We noticed the following weird behavior after using Auto from the
> Brightness/Contrast Tool using a non-rectangular ROI.
> The pixels inside the rectangle containing the ROI but outside the ROI
> itself get modified in an unexpected way. The result is quite
> beautiful but not really wanted. The problem seems to occur only if
> the image was inverted just before.
> To reproduce the problem:
> - File/New/Image...
> create image filled with Ramp
> - Edit/Invert
> - make any non-rectangular selection
> - press the 'Auto' button in the Brightness/Contrast tool
> The problem does not occur with
> Process/Enhance Contrast

This bug is fixed in the v1.41c daily build.

-wayne