[9fans] shrinking an Image?

[9fans] shrinking an Image?

[Axel Belinfante]

Is there a good/easy/general way to shrink an Image?
(read 'arbitrary' image, shrink it, then use shrunken Image)
in this particular case: to a quarter of its size (48x48 -> 24x24),
where the Image is read from a face file
(I guess it amounts to computing a single pixel
  for every four of the original, but how?)

Axel. - trying to improve a bitsy sized rushhour, looks like
touch screens are easier for small children than mice

Re: [9fans] shrinking an Image?

[andrey mirtchovski]

On 8/26/06, Axel Belinfante <Axel.Belinfante@cs.utwente.nl> wrote:
> Is there a good/easy/general way to shrink an Image?

would resample(1) do?

http://swtch.com/plan9port/man/man1/resample.html

Re: [9fans] shrinking an Image?

[Federico Benavento]

resample(1)

On 8/26/06, Axel Belinfante <Axel.Belinfante@cs.utwente.nl> wrote:
> Is there a good/easy/general way to shrink an Image?
> (read 'arbitrary' image, shrink it, then use shrunken Image)
> in this particular case: to a quarter of its size (48x48 -> 24x24),
> where the Image is read from a face file
> (I guess it amounts to computing a single pixel
>   for every four of the original, but how?)
>
> Axel. - trying to improve a bitsy sized rushhour, looks like
> touch screens are easier for small children than mice
>


-- 
Federico G. Benavento

Re: [9fans] shrinking an Image?

[lucio]

> would resample(1) do?

I find that resample(1) is not visually as faithful as (even?)
page(1)'s built-in resizer.  I have not explored the possible causes,
however.  For JPEG and 2, 4, 8 times resizing (shrinking) I find the
cjpeg | djpeg combination to be preferable.

++L

Re: [9fans] shrinking an Image?

[Russ Cox]

> I find that resample(1) is not visually as faithful as (even?)
> page(1)'s built-in resizer.  I have not explored the possible causes,
> however.

They should behave identically.
Page's zoom function uses a copy of the resample code.

Russ

Re: [9fans] shrinking an Image?

[erik quanstrom]

in light of russ' comment, could it be that you are not using the -t
option for jpg?  the default for jpg/png/etc is to downsample to 8 bits.
page actually uses "-t9" as it's options.  i've been burned by this 
many times.  perhaps jpg should default to depth 24 when being
used as a filter.

- erik

On Sun Aug 27 02:55:23 CDT 2006, lucio@proxima.alt.za wrote:
> > would resample(1) do?
> 
> I find that resample(1) is not visually as faithful as (even?)
> page(1)'s built-in resizer.  I have not explored the possible causes,
> however.  For JPEG and 2, 4, 8 times resizing (shrinking) I find the
> cjpeg | djpeg combination to be preferable.

Re: [9fans] shrinking an Image?

[lucio]

> They should behave identically.
> Page's zoom function uses a copy of the resample code.

It was a long time ago (maybe a couple of years) that I last used
resample(1) and the results were disappointing.  I have not checked
whether it has been revamped recently.  And perhaps I was using it
sub-optimally, as I was almost certainly applying it to GIFs or JPEGs.

I do recall noting that page(1) did a more faithful job.  If memory
serves, I may be able to try it again on the same image, but I make no
promises.

++L

Re: [9fans] shrinking an Image?

[lucio]

> in light of russ' comment, could it be that you are not using the -t
> option for jpg?  the default for jpg/png/etc is to downsample to 8 bits.
> page actually uses "-t9" as it's options.  i've been burned by this 
> many times.  perhaps jpg should default to depth 24 when being
> used as a filter.

Could be, I did not spend much time on it.

++L

Re: [9fans] shrinking an Image?

[Axel Belinfante]

> > Is there a good/easy/general way to shrink an Image?
> 
> would resample(1) do?

thanks, I succeeded to overlook it.

resample(1) contains a reference to crop(1);
it might be nice to add a reverse ref in crop(1)

Re: [9fans] shrinking an Image?

[Axel Belinfante]

Now that you mention this, for one plan 9 image that I converted
using togif and tojpg I had better results after first converting
it using 'iconv -c m8'.
Was I 'unlucky' in needing this additional step, or is that
the usual procedure?

Axel.

> in light of russ' comment, could it be that you are not using the -t
> option for jpg?  the default for jpg/png/etc is to downsample to 8 bits.
> page actually uses "-t9" as it's options.  i've been burned by this 
> many times.  perhaps jpg should default to depth 24 when being
> used as a filter.
> 
> - erik
> 
> On Sun Aug 27 02:55:23 CDT 2006, lucio@proxima.alt.za wrote:
> > > would resample(1) do?
> > 
> > I find that resample(1) is not visually as faithful as (even?)
> > page(1)'s built-in resizer.  I have not explored the possible causes,
> > however.  For JPEG and 2, 4, 8 times resizing (shrinking) I find the
> > cjpeg | djpeg combination to be preferable.

Re: Re: [9fans] shrinking an Image?

[andrey mirtchovski]

i seem to remember tojpg and friends doing a floyd-steinberg error
diffusion to downsample to 8bpp by default.

see the difference in these two (rather old) screenshots, the first
taken with topng, the second most likely from a running drawterm:

http://pages.cpsc.ucalgary.ca/~mirtchov/screenshots/scr-plan9.png

http://pages.cpsc.ucalgary.ca/~mirtchov/screenshots/scr-1.png

the acme titles experience the most obvious deviation.

Re: [9fans] shrinking an Image?

[erik quanstrom]

you could have skipped this step by disabling error diffusion.
the bugs section of iconv(1) mentions this.

i think the main problem with the current error diffusion is that
the scan always goes from left to right.  a simple fix might be to
alternate the scan from left-to-right then right-to-left.  also, the
error diffusion needs to be extended beyond 3 color channels
(for example, to b&w).

perhaps i don't understand the problem properly, but i think it
might make sense to remove the error diffusion from jpg and friends
and add it to iconv.

- erik

Re: [9fans] shrinking an Image?

[Steve Simon]

> I find that resample(1) is not visually as faithful as (even?)
> page(1)'s built-in resizer.  I have not explored the possible causes,
> however.  For JPEG and 2, 4, 8 times resizing (shrinking) I find the
> cjpeg | djpeg combination to be preferable.

Just a guess, but prehaps page does resample in screen space 
(e.g. 16bpp) whereas resample(1) probably runs faithfully in YUV
(well, more likely RGB) space. This could easily have the effect of
amplifying the JPEG compression artiacts. Resampling in the incorrect
gamma land can also have strange effects like this.

-Steve

Re: [9fans] shrinking an Image?

[erik quanstrom]

the algorithm is exactly the same.  both convert the image to 8 bits-per-channel then resample
using a kaiser window.  no 16 bit channel or alternate colorspaces anywhere.

- erik