phatch-dev team mailing list archive

[Stani <spe.stani.be@xxxxxxxxx>]

There is an interesting thread currently at the image-sig mailing list
about downsampling images. The methods suggested by Douglas and Franz
might be interesting techniques to add to Phatch or is it just
clutter? What do you think?

---------- Forwarded message ----------
From: Douglas Bagnall <douglas@xxxxxxxxxxxxxxx>
Date: Wed, Jun 3, 2009 at 3:57 AM
Subject: Re: [Image-SIG] Lanczos interpolation
To: image-sig@xxxxxxxxxx



If you want the downsampling algorithm used by the Gimp and (I think)
Photoshop, try:

def stretch(im, size, filter=Image.NEAREST):
   im.load()
   im = im._new(im.im.stretch(size, filter))
   return im

In 2005 I found it to be a few times quicker than ANTIALIAS resizing,
and the results were better from what I was doing.

(From http://mail.python.org/pipermail/image-sig/2005-May/003310.html)


Douglas
_______________________________________________
Image-SIG maillist  -  Image-SIG@xxxxxxxxxx
http://mail.python.org/mailman/listinfo/image-sig
>
> On Sun, May 10, 2009 at 12:45 PM, David Yan <davidyan6@xxxxxxxxx> wrote:
> > I want to resize an image Lanczos interpolation but I haven't found any
> > python module that can do so.  Also, I'm not knowledgeable enough to write
> > one myself.  What's the best way I can do this?  I use PIL currently but the
> > resampling filters don't suit my needs (antialias is too slow, bilinear and
> > bicubic produce aliasing when downsampling).
>
> ANTIALIAS is a Lanczos interpolation written in C, so it's not
> entirely obvious to me how you expect a Python version to be faster...
>
> </F>


---------- Forwarded message ----------
From: Franz Buchinger <fbuchinger@xxxxxxxxx>
Date: Tue, Jun 2, 2009 at 1:58 PM
Subject: Re: [Image-SIG] Lanczos interpolation
To: image-sig@xxxxxxxxxx


A good old trick is to work with intermediate images: resize the image
using the Bilinear method and produce an intermediate image that is
about 25% larger than the final image.
Then you can downscale the intermediate image using Lanczos.

This should give a much better performance than working just with
Lanczos, especially for large downscaling ratios (e.g. 10 Megapixel
JPEG -> 1024x768) .
Image quality shouldn't be affected too much by this method.


Franz

-- 
Phatch Photo Batch Processor - http://photobatch.stani.be
SPE Python IDE - http://pythonide.stani.be