From Richard.E.Brown at DARTWARE.COM  Mon Sep  3 09:03:01 2007
From: Richard.E.Brown at DARTWARE.COM (Richard E. Brown)
Date: 03 Sep 2007 03:03:01 -0400
Subject: [Image-SIG] Trouble building PIL 1.1.6 on MacOS X Intel
Message-ID: <3225826@blitz.dartware.com>

An embedded and charset-unspecified text was scrubbed...
Name: not available
Url: http://mail.python.org/pipermail/image-sig/attachments/20070903/25d7dfd4/attachment.txt 

From tim at concussion.cc  Wed Sep  5 22:34:31 2007
From: tim at concussion.cc (Concussion GFX - Tim)
Date: Wed, 05 Sep 2007 16:34:31 -0400
Subject: [Image-SIG] PSD support help
Message-ID: <46DF12D7.5070801@concussion.cc>

hey list!

Is there any info on an updated plugin for PSD's (greater then version 3?)

I have been playing around with PSD's in pil for the last couple of days 
and it has been hit and miss with getting PSD files to load. After 
digging a little deeper it turns out that if the bottom most layer 
(layer 0 i presume) has transparency, then it will not load the PSD 
file. If I create a new layer with a solid fill and move it to the 
bottom,save the PSD, then the PSD will load.

If I comment out the following line, the PSD will load, but the channels 
are not lined up correctly.
Also if there is an extra channel in the file, i get the same 'not 
lined-up' error in the rendered image.

self.layers = _layerinfo(self.fp)

Have a look at this jpg to see how the PSD are being displayed.
http://www.concussion.cc/preview.jpg

Thanks
-Tim

From jantod at gmail.com  Thu Sep  6 17:09:40 2007
From: jantod at gmail.com (Janto Dreijer)
Date: Thu, 6 Sep 2007 17:09:40 +0200
Subject: [Image-SIG] more descriptive error for image.paste
Message-ID: <cbc7eb430709060809ie16f305o9fa3c244f1fd0508@mail.gmail.com>

I'd really appreciate it if Image.paste can be more explicit about
what went wrong than just "ValueError: images do not match".

I've been sitting here for at least an hour trying to figure out why
the images are incompatible. Different color mode? Pasted image size
!= bbox size? bbox outside image? Wrong parmeters? Argh! Just tell me!

Thanks
Janto

From premal at crestgarments.com  Mon Sep 10 17:20:19 2007
From: premal at crestgarments.com (premal at crestgarments.com)
Date: Mon, 10 Sep 2007 19:20:19 +0400
Subject: [Image-SIG] **VIRUS**  Hi
Message-ID: <142D05C69@tydex.ru>

The original message was received at Mon, 10 Sep 2007 19:20:19 +0400 from 163.86.100.134

----- The following addresses had permanent fatal errors -----
image-sig at python.org

----- Transcript of session follows -----
... while talking to host 207.189.228.24:
>>> RCPT To:<image-sig at python.org>
<<< 550 MAILBOX NOT FOUND

-------------- next part --------------
Kerio WinRoute Firewall email scanner found a virus in the following attachment:
	Name:	instruction.zip
	Content type:	application/octet-stream
Additional information from antivirus: External AV verdict: Win32/Mydoom.R worm
	
The attachment has been removed.


From gandalf at shopzeus.com  Tue Sep 11 13:45:53 2007
From: gandalf at shopzeus.com (Laszlo Nagy)
Date: Tue, 11 Sep 2007 13:45:53 +0200
Subject: [Image-SIG] Antialiased text on transparent image fails?
Message-ID: <46E67FF1.70705@shopzeus.com>


 Hi All,

I tried to put a text on a transparent image. Here is a test:

import Image
import ImageDraw
import ImageFont

img = Image.new('RGBA',(100,20),(255,0,0,0))
drawer = ImageDraw.Draw(img)
fnt = ImageFont.truetype("Vera.ttf",20) # 
http://ftp.gnome.org/pub/GNOME/sources/ttf-bitstream-vera/1.10/
drawer.text((0,0),"ABCDE",font=fnt,fill="#00ff00")
img.save("test.png") # Result image

I'm working on a rendering engine but I created this example to show the 
problem. The rendering engine should be able to create an image with 
transparent parts, and one should be able to put the rendered image on 
top of any other image.

The problem itself: font edges are interpolated between the background 
and foreground color. You can see it on the resulting image. If you open 
the created "test.png" file in GIMP and use the color picker tool then 
you can see values like:

(203,52,0,52)
(215,40,0,40)


I think that this is bad. The background was fully transparent. If you 
put a green object on a fully transparent thing, you should never see 
any red in it. I believe that the result should be something like

(0,52,0,52)
(0,40,0,40)

In other words, when antialiasing a text, the background pixel's color 
should be weighted with its transparency. In my example, the background 
pixel is fully red but should have zero weight.

Workarounds?

Working with black or white initial background is not a workaround, 
because PIL will darken/lighten the pixels. I used red+green just to 
make the problem more visible. A correct workaround is to use the actual 
target background that will finally be used, but it is not a good 
workaround. This is obvious: I want to render the result image once, 
then put it on different target images. Rendering the result image each 
time I need to put it on a target would be very slow.

Can you please confirm if this is a bug in PIL?  Comments welcome.

Best,

   Laszlo

-------------- next part --------------
A non-text attachment was scrubbed...
Name: test.png
Type: image/png
Size: 1524 bytes
Desc: not available
Url : http://mail.python.org/pipermail/image-sig/attachments/20070911/29ca4ada/attachment.png 

From cyril.giraudon at free.fr  Tue Sep 11 17:26:25 2007
From: cyril.giraudon at free.fr (cyril giraudon)
Date: Tue, 11 Sep 2007 17:26:25 +0200
Subject: [Image-SIG] PIL Pixmap from XLIB Pixmap
Message-ID: <46E6B3A1.2020704@free.fr>

Hi,

I 'm currently trying to write a xulrunner windowless plugin in python.
Data are drawn in a xlib pixmap (a XID).

I 'd like to know if it's possible to create a PIL image from XLIB
Pixmap XID ?
And then expose the image into this pximap.

Under windows, plugins uses an HDC. PIL can expose images in HDC object,
I 'd like the same behaviour under UNIX.

Thanks a lot,

Cyril.


From justin2 at fagnani.com  Wed Sep 12 01:04:14 2007
From: justin2 at fagnani.com (Justin Fagnani-Bell)
Date: Tue, 11 Sep 2007 16:04:14 -0700
Subject: [Image-SIG] Antialias, Bicubic resampling not working...
Message-ID: <3CAE6C6A-E82E-49FF-BC3E-A1E953170CF4@fagnani.com>

I'm writing a thumb-nailing script that also rotates the thumbs, so  
I'm using both Image.thumbnail and Image.rotate. Neither antialiasing  
with thumbnail, or bicubic with rotate are working, adn the results  
look horrendous with nearest neighbor. I'm not getting any errors,  
and I haven't figured anything out by digging through the PIL source.  
I have version 1.1.6 installed via MacPorts.

Any ideas?

Thanks,
   Justin


From cyril.giraudon at free.fr  Wed Sep 12 10:11:31 2007
From: cyril.giraudon at free.fr (cyril giraudon)
Date: Wed, 12 Sep 2007 10:11:31 +0200
Subject: [Image-SIG] PIL Pixmap from XLIB Pixmap
In-Reply-To: <46E6B3A1.2020704@free.fr>
References: <46E6B3A1.2020704@free.fr>
Message-ID: <46E79F33.7050001@free.fr>

Sorry, I wanted to say :
I 'd like to know if it's possible to expose a PIL image into an XLIB
Pixmap known by its XID ?
(create a PIL Pixmap from an XID).

Cyril.

cyril giraudon a ?crit :
> Hi,
> 
> I 'm currently trying to write a xulrunner windowless plugin in python.
> Data are drawn in a xlib pixmap (a XID).
> 
> I 'd like to know if it's possible to create a PIL image from XLIB
> Pixmap XID ?
> And then expose the image into this pximap.
> 
> Under windows, plugins uses an HDC. PIL can expose images in HDC object,
> I 'd like the same behaviour under UNIX.
> 
> Thanks a lot,
> 
> Cyril.
> 
> _______________________________________________
> Image-SIG maillist  -  Image-SIG at python.org
> http://mail.python.org/mailman/listinfo/image-sig
> 


From connellybarnes at yahoo.com  Thu Sep 13 07:57:02 2007
From: connellybarnes at yahoo.com (Connelly Barnes)
Date: Wed, 12 Sep 2007 22:57:02 -0700 (PDT)
Subject: [Image-SIG] Filter PIL images with arbitrary kernels via numpy and
	FFT
Message-ID: <284800.85171.qm@web54304.mail.re2.yahoo.com>


The Python Imaging Library currently only supports 3x3 and 5x5 image filters.  Here's a module
which uses numpy and the FFT to compute filters of arbitrary size:

  http://barnesc.blogspot.com/2007/09/filter-numpy-images-with-fft-python.html

For an image and filter which contain a total of N pixels overall, it runs in O(N*log(N)) time;
however, numpy's FFT isn't too fast and so in practice other methods may be more appropriate to
compute filters quickly, as described in the blog post linked above.

Connelly Barnes
http://www.connellybarnes.com/



       
____________________________________________________________________________________
Looking for a deal? Find great prices on flights and hotels with Yahoo! FareChase.
http://farechase.yahoo.com/

From coen at reservoir.nl  Fri Sep 14 15:36:00 2007
From: coen at reservoir.nl (Coen van der Kamp)
Date: Fri, 14 Sep 2007 15:36:00 +0200
Subject: [Image-SIG] Antialiased text on transparent image fails?
In-Reply-To: <46E67FF1.70705@shopzeus.com>
References: <46E67FF1.70705@shopzeus.com>
Message-ID: <46EA8E40.8050307@reservoir.nl>

Hi Laszlo,

Isn't this the way it should work? The alpha value of a pixel is treated 
the same way as any other channel. When a pixel is half covered by a 
shape, PIL wil draw 50% of the foreground and 50% of the background. In 
your case: a green shape 0, 255, 0, 255 (rgba) and background 255, 0, 0, 
0. If you take the average of the two you come up with R=128 in the 
pixel and will be visible for 50% A=128. Isn't this the desired effect? 
And if it would be a color? Don't you want the same to happen?

The solution? Render the shape on a 0,0,0,0 background. No red will be 
added, your half filled pixel with a correct alpha value will blend with 
new backgrounds as it is supposed to. You can save this image in various 
ways for later use.

Just my thought... and i hope it helps.
Greetings,

Coen


Laszlo Nagy wrote:
>
> Hi All,
>
> I tried to put a text on a transparent image. Here is a test:
>
> import Image
> import ImageDraw
> import ImageFont
>
> img = Image.new('RGBA',(100,20),(255,0,0,0))
> drawer = ImageDraw.Draw(img)
> fnt = ImageFont.truetype("Vera.ttf",20) # 
> http://ftp.gnome.org/pub/GNOME/sources/ttf-bitstream-vera/1.10/
> drawer.text((0,0),"ABCDE",font=fnt,fill="#00ff00")
> img.save("test.png") # Result image
>
> I'm working on a rendering engine but I created this example to show 
> the problem. The rendering engine should be able to create an image 
> with transparent parts, and one should be able to put the rendered 
> image on top of any other image.
>
> The problem itself: font edges are interpolated between the background 
> and foreground color. You can see it on the resulting image. If you 
> open the created "test.png" file in GIMP and use the color picker tool 
> then you can see values like:
>
> (203,52,0,52)
> (215,40,0,40)
>
>
> I think that this is bad. The background was fully transparent. If you 
> put a green object on a fully transparent thing, you should never see 
> any red in it. I believe that the result should be something like
>
> (0,52,0,52)
> (0,40,0,40)
>
> In other words, when antialiasing a text, the background pixel's color 
> should be weighted with its transparency. In my example, the 
> background pixel is fully red but should have zero weight.
>
> Workarounds?
>
> Working with black or white initial background is not a workaround, 
> because PIL will darken/lighten the pixels. I used red+green just to 
> make the problem more visible. A correct workaround is to use the 
> actual target background that will finally be used, but it is not a 
> good workaround. This is obvious: I want to render the result image 
> once, then put it on different target images. Rendering the result 
> image each time I need to put it on a target would be very slow.
>
> Can you please confirm if this is a bug in PIL?  Comments welcome.
>
> Best,
>
>   Laszlo
>
>
> ------------------------------------------------------------------------
>
> ------------------------------------------------------------------------
>
> _______________________________________________
> Image-SIG maillist  -  Image-SIG at python.org
> http://mail.python.org/mailman/listinfo/image-sig
>   


From btvlinux at orange.nl  Sun Sep 16 14:36:51 2007
From: btvlinux at orange.nl (Bauke Veenstra)
Date: Sun, 16 Sep 2007 14:36:51 +0200
Subject: [Image-SIG] PSDraw instance has no attribute 'textsize'
Message-ID: <1189946211.6008.0.camel@localhost.localdomain>

Running the example from the excellent PIL handbook gives an error, 
after these two lines: 
ps.setfont("HelveticaNarrow-Bold", 36) 
w, h, b = ps.textsize(title) 
AttributeError: PSDraw instance has no attribute 'textsize' 


This error occurred on my Linux and on my Windows system. 


Regards, 
Bauke 




From bloominator at hotmail.com  Fri Sep 14 04:50:39 2007
From: bloominator at hotmail.com (Gary Bloom)
Date: Thu, 13 Sep 2007 22:50:39 -0400
Subject: [Image-SIG] PIL DPI trouble
Message-ID: <BAY113-DAV1B4F4E6832AFBD8872083A3BC0@phx.gbl>

Howdy!

I notice that the PIL has trouble playing with the DPI of other programs and image standards: when I create a file with Paint Shop Pro or Photoshop, the PIL routines can't properly recognize the DPI.  And when I create an image file (via the save method) of PIL, setting the DPI, those same two programs can't read the DPI as set by them.  Is there a workaround for this?

Other than this one issue, I have found the PIL to be a BLAST!  :)

Thanks!

Gary Bloom
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.python.org/pipermail/image-sig/attachments/20070913/84ead2b5/attachment.htm 

From e.man.69 at gmail.com  Fri Sep 14 18:24:27 2007
From: e.man.69 at gmail.com (E.Masta)
Date: Fri, 14 Sep 2007 12:24:27 -0400
Subject: [Image-SIG] Installation not succesful on Vista
Message-ID: <8bc5ff710709140924p2f424669ode1e6a250875b68@mail.gmail.com>

When I try to install, it spews some error messages. Can you tell me what to
do?

Thanks.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.python.org/pipermail/image-sig/attachments/20070914/4d66b450/attachment.htm 

From fredrik at pythonware.com  Mon Sep 17 00:36:28 2007
From: fredrik at pythonware.com (Fredrik Lundh)
Date: Mon, 17 Sep 2007 00:36:28 +0200
Subject: [Image-SIG] 1.1.5 and 1.1.6 compatibility issues
In-Reply-To: <469B6484.8020105@redhat.com>
References: <469B6484.8020105@redhat.com>
Message-ID: <fckb5g$3et$1@sea.gmane.org>

Joel Andres Granados wrote:

> Is version 1.1.6 backward compatible with 1.1.5?

Yes, with the possible exception of code that relies on bugs or 
undefined behaviour.  See the CHANGES document in the source 
distribution for more details.

</F>


From fredrik at pythonware.com  Mon Sep 17 00:46:32 2007
From: fredrik at pythonware.com (Fredrik Lundh)
Date: Mon, 17 Sep 2007 00:46:32 +0200
Subject: [Image-SIG] Installation not succesful on Vista
In-Reply-To: <8bc5ff710709140924p2f424669ode1e6a250875b68@mail.gmail.com>
References: <8bc5ff710709140924p2f424669ode1e6a250875b68@mail.gmail.com>
Message-ID: <fckbod$4kb$1@sea.gmane.org>

E.Masta wrote:

> When I try to install, it spews some error messages. Can you tell me 
> what to do?

Telling us what the error messages are might help.

</F>


From fredrik at pythonware.com  Mon Sep 17 00:47:56 2007
From: fredrik at pythonware.com (Fredrik Lundh)
Date: Mon, 17 Sep 2007 00:47:56 +0200
Subject: [Image-SIG] Antialias, Bicubic resampling not working...
In-Reply-To: <3CAE6C6A-E82E-49FF-BC3E-A1E953170CF4@fagnani.com>
References: <3CAE6C6A-E82E-49FF-BC3E-A1E953170CF4@fagnani.com>
Message-ID: <fckbr0$4kb$2@sea.gmane.org>

Justin Fagnani-Bell wrote:

> I'm writing a thumb-nailing script that also rotates the thumbs, so  
> I'm using both Image.thumbnail and Image.rotate. Neither antialiasing  
> with thumbnail, or bicubic with rotate are working, adn the results  
> look horrendous with nearest neighbor. I'm not getting any errors,  
> and I haven't figured anything out by digging through the PIL source.  
> I have version 1.1.6 installed via MacPorts.

What's the "mode" attribute for these images?  Does it work better if you do

    image = image.convert("RGB")

first?

</F>


From lucadex at gmail.com  Mon Sep 17 09:36:16 2007
From: lucadex at gmail.com (Luca De Santis)
Date: Mon, 17 Sep 2007 09:36:16 +0200
Subject: [Image-SIG] PIL DPI trouble
In-Reply-To: <BAY113-DAV1B4F4E6832AFBD8872083A3BC0@phx.gbl>
References: <BAY113-DAV1B4F4E6832AFBD8872083A3BC0@phx.gbl>
Message-ID: <2ce6cd010709170036x29666383rff5f5a7c9c2a164b@mail.gmail.com>

That's exactly the same problem I incurred into.
I tried PIL  version 1.1.6 on a Mac OS X 10.4 and on an old Linux box
and can't get the DPI attribute for JPEG and PNG files that I've
created with Photoshop.

I tested that both on Python 2.5 and Python 2.1.3 with the same result.

I tried also to update libjpeg on Linux but nothing changes.

Any help on that is greatly appreciated.
TIA.

Ciao,
Luca

On 9/14/07, Gary Bloom <bloominator at hotmail.com> wrote:
> I notice that the PIL has trouble playing with the DPI of other programs and
> image standards: when I create a file with Paint Shop Pro or Photoshop, the
> PIL routines can't properly recognize the DPI.  And when I create an image
> file (via the save method) of PIL, setting the DPI, those same two programs
> can't read the DPI as set by them.  Is there a workaround for this?
>
> Other than this one issue, I have found the PIL to be a BLAST!  :)
>
> Thanks!
>
> Gary Bloom

-- 
Luca De Santis
Pisa, Italy

From justin2 at fagnani.com  Mon Sep 17 19:10:32 2007
From: justin2 at fagnani.com (Justin Fagnani-Bell)
Date: Mon, 17 Sep 2007 10:10:32 -0700
Subject: [Image-SIG] Antialias, Bicubic resampling not working...
In-Reply-To: <fckbr0$4kb$2@sea.gmane.org>
References: <3CAE6C6A-E82E-49FF-BC3E-A1E953170CF4@fagnani.com>
	<fckbr0$4kb$2@sea.gmane.org>
Message-ID: <21CAAD36-53F6-446E-A010-ED863F8E3A30@fagnani.com>

So the mode was P and converting to RGB fixed the thumbnail issue,  
but it didn't fix the rotate problem. It very noticeable, because I'm  
using expand=True with a white background test image, and the edges  
are very jagged.
I'm saving as a PNG, and there's pretty much nothing going on other  
than a thumbnail and rotate.

Any other ideas?

Thanks a lot,
   Justin


On Sep 16, 2007, at 3:47 PM, Fredrik Lundh wrote:

> Justin Fagnani-Bell wrote:
>
>> I'm writing a thumb-nailing script that also rotates the thumbs, so
>> I'm using both Image.thumbnail and Image.rotate. Neither antialiasing
>> with thumbnail, or bicubic with rotate are working, adn the results
>> look horrendous with nearest neighbor. I'm not getting any errors,
>> and I haven't figured anything out by digging through the PIL source.
>> I have version 1.1.6 installed via MacPorts.
>
> What's the "mode" attribute for these images?  Does it work better  
> if you do
>
>     image = image.convert("RGB")
>
> first?
>
> </F>
>
> _______________________________________________
> Image-SIG maillist  -  Image-SIG at python.org
> http://mail.python.org/mailman/listinfo/image-sig


From lucadex at gmail.com  Fri Sep 21 12:12:24 2007
From: lucadex at gmail.com (Luca De Santis)
Date: Fri, 21 Sep 2007 12:12:24 +0200
Subject: [Image-SIG] PIL DPI trouble - Solved
Message-ID: <2ce6cd010709210312u7b7d474ct48bb877d7eb0c09b@mail.gmail.com>

I found out that the behaviour of PIL was totally correct and simply
Photoshop "adds" some more information sometimes when a JPEG file is
opened.

I summarize what I discovered here, hoping it might be useful to others.

The JPEG standard doesn't cover a way to encode the resolution of a
file but the JFIF extension does. For a quick and useful introduction
to these concepts you can visit the Wikipedia page
http://en.wikipedia.org/wiki/JFIF
Here you can find a very useful table with the byte map of a JFIF file.

In particular the "Density Units" field is a byte with three possible values:
    * 0 - No units, aspect ratio only specified
    * 1 - Pixels per Inch
    * 2 - Pixels per Centimetre
With a value of Density Units == 1 or == 2 you can get the resolution
information in the next two fields (2 bytes long each), that is "X
Density" and "Y Density". Of course these map exactly "dpi" when
Density Units == 1.

For all the images that I opened with PIL and couldn't find the dpi
attribute I "hex-dumped" them and found out the following values for
these fields:
    * Density Unit: 0
    * X Density: 1
    * Y Density: 1

I assume that in these cases Photoshop simply sets the resolution to
the default for screens (which is 72 dpi at least on the Mac).
Correctly PIL doesn't set the dpi attribute in the "info" dictionary.

To cut this long story short... I was wrong and PIL rocks hard! :-)

Hope my experience could help others.

Ciao,
Luca

Luca De Santis
Pisa, Italy

On 9/17/07, Luca De Santis <lucadex at gmail.com> wrote:
> That's exactly the same problem I incurred into.
> I tried PIL  version 1.1.6 on a Mac OS X 10.4 and on an old Linux box
> and can't get the DPI attribute for JPEG and PNG files that I've
> created with Photoshop.
>
> I tested that both on Python 2.5 and Python 2.1.3 with the same result.
>
> I tried also to update libjpeg on Linux but nothing changes.
>
> Any help on that is greatly appreciated.
> TIA.

From bloominator at hotmail.com  Fri Sep 21 21:28:57 2007
From: bloominator at hotmail.com (Gary Bloom)
Date: Fri, 21 Sep 2007 15:28:57 -0400
Subject: [Image-SIG] PIL DPI trouble - Solved
References: <2ce6cd010709210312u7b7d474ct48bb877d7eb0c09b@mail.gmail.com>
Message-ID: <BAY113-DAV15FCB3D9A98A13FAB6984BA3BB0@phx.gbl>

Hi Luca,

Thanks for shedding some light on this topic.  However, what you've said has 
left me even more curious than before.  There are still some things I don't 
understand...

For one, why do Photoshop (PS) and Paint Shop Pro (PSP) play so nicely 
together when the PIL doesn't play along at all for DPI.  If I create an 
image with certain DPI in Photoshop, Paint Shop Pro can read the image and 
see that set density, and vice versa.  But the PIL doesn't recognize the DPI 
of either a PSP or Photoshop image.  And of course neither PS or PSP 
recognize the DPI of a PIL image.  So I'm a bit confused.  You say that PIL 
seems to be "playing by the rules" and adhering to the JFIF standard.  (My 
images are both TIFF and JPEG, but I'm guessing that it's the same problem 
regardless of which image type it is.), yet PSP and PS cannot read the DPI 
of these PIL-processed images.  This doesn't make sense to me.  If, what you 
say below is true and PSP and PS are adding "extra stuff" outside of the 
JFIF standard, is it coincidental that they can understand one another? 
And, even if they are both writing (and reading the other's) non-standard 
DPI info, shouldn't the PIL be foresighted enough to have a "compatibility 
mode"' for that "extra stuff"?  Sometimes just adhering blindly to a 
standard is a good thing, unless the rest of the world is ignoring the 
standard, which is what it seems PSP and PS are doing.  So do you have any 
idea how to make the PIL compatible with these other programs?  At work my 
company has a product which also deals with images, and one day I imported 
an image that I had resized and changed the DPI of it via the PIL, and it 
made our software crash.  This has never happened when using any other 
program to manipulate images.  Any thoughts on this?  Or perhaps can you 
point me to someone who can shed even more light on the matter?

Thanks so much for your reply,

Gary
----- Original Message ----- 
From: "Luca De Santis" <lucadex at gmail.com>
To: <Image-SIG at python.org>
Cc: <bloominator at hotmail.com>
Sent: Friday, September 21, 2007 6:12 AM
Subject: PIL DPI trouble - Solved


>I found out that the behaviour of PIL was totally correct and simply
> Photoshop "adds" some more information sometimes when a JPEG file is
> opened.
>
> I summarize what I discovered here, hoping it might be useful to others.
>
> The JPEG standard doesn't cover a way to encode the resolution of a
> file but the JFIF extension does. For a quick and useful introduction
> to these concepts you can visit the Wikipedia page
> http://en.wikipedia.org/wiki/JFIF
> Here you can find a very useful table with the byte map of a JFIF file.
>
> In particular the "Density Units" field is a byte with three possible 
> values:
>    * 0 - No units, aspect ratio only specified
>    * 1 - Pixels per Inch
>    * 2 - Pixels per Centimetre
> With a value of Density Units == 1 or == 2 you can get the resolution
> information in the next two fields (2 bytes long each), that is "X
> Density" and "Y Density". Of course these map exactly "dpi" when
> Density Units == 1.
>
> For all the images that I opened with PIL and couldn't find the dpi
> attribute I "hex-dumped" them and found out the following values for
> these fields:
>    * Density Unit: 0
>    * X Density: 1
>    * Y Density: 1
>
> I assume that in these cases Photoshop simply sets the resolution to
> the default for screens (which is 72 dpi at least on the Mac).
> Correctly PIL doesn't set the dpi attribute in the "info" dictionary.
>
> To cut this long story short... I was wrong and PIL rocks hard! :-)
>
> Hope my experience could help others.
>
> Ciao,
> Luca
>
> Luca De Santis
> Pisa, Italy
>
> On 9/17/07, Luca De Santis <lucadex at gmail.com> wrote:
>> That's exactly the same problem I incurred into.
>> I tried PIL  version 1.1.6 on a Mac OS X 10.4 and on an old Linux box
>> and can't get the DPI attribute for JPEG and PNG files that I've
>> created with Photoshop.
>>
>> I tested that both on Python 2.5 and Python 2.1.3 with the same result.
>>
>> I tried also to update libjpeg on Linux but nothing changes.
>>
>> Any help on that is greatly appreciated.
>> TIA.
> 


From jcupitt at gmail.com  Sat Sep 22 12:51:23 2007
From: jcupitt at gmail.com (jcupitt at gmail.com)
Date: Sat, 22 Sep 2007 11:51:23 +0100
Subject: [Image-SIG] PIL DPI trouble - Solved
In-Reply-To: <2ce6cd010709210312u7b7d474ct48bb877d7eb0c09b@mail.gmail.com>
References: <2ce6cd010709210312u7b7d474ct48bb877d7eb0c09b@mail.gmail.com>
Message-ID: <522c6460709220351q7e95fd04s702aaaec5a4dc27d@mail.gmail.com>

On 9/21/07, Luca De Santis <lucadex at gmail.com> wrote:
> The JPEG standard doesn't cover a way to encode the resolution of a
> file but the JFIF extension does. For a quick and useful introduction
> to these concepts you can visit the Wikipedia page
> http://en.wikipedia.org/wiki/JFIF
> Here you can find a very useful table with the byte map of a JFIF file.

My understanding is that EXIF has now pretty much replaced the JFIF
resolution system.

http://en.wikipedia.org/wiki/Exif

I think most image processing systems use EXIF to put DPI information
into JPEG images. PIL will read EXIF tags from a JPEG image, but (as
far as I can see) will not write tags back again. Googling for "PIL
EXIF" seems uncertain anyway.

John

From khademi.somayeh63 at gmail.com  Sun Sep 23 09:37:37 2007
From: khademi.somayeh63 at gmail.com (somayeh khademi)
Date: Sun, 23 Sep 2007 11:07:37 +0330
Subject: [Image-SIG] screenshot in python
Message-ID: <7e7b076d0709230037t2b621f1bjd0a755a1103db850@mail.gmail.com>

Hello,



I want write a screenshot with python, Has python any module related it?

 please help me about it.



Thank for future help.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.python.org/pipermail/image-sig/attachments/20070923/2b659739/attachment.htm 

From douglas at paradise.net.nz  Mon Sep 24 03:48:49 2007
From: douglas at paradise.net.nz (Douglas Bagnall)
Date: Mon, 24 Sep 2007 13:48:49 +1200
Subject: [Image-SIG] crop is a little too lazy (with patch)
Message-ID: <46F71781.60806@paradise.net.nz>


A cropped image does not return a pixel access object when load()ed.
This seems to be a simple oversight.

  >>> im = Image.new('L',(50,50))
  >>> im.load()
  <PixelAccess object at 0xb7d900f0>
  >>> im2 = im.crop((0, 0, 20, 20))
  >>> im2.load()
  None


There is an easy workaround:

  >>> im2.im.pixel_access()
  <PixelAccess object at 0xb7d900f0>

and the patch below cures the problem.

douglas

---------------------------------------------

--- Image-old.py        2007-09-24 13:19:30.000000000 +1200
+++ Image-new.py        2007-09-24 13:19:17.000000000 +1200
@@ -1666,6 +1666,8 @@
         if self.__crop:
             self.im = self.im.crop(self.__crop)
             self.__crop = None
+        if self.im:
+            return self.im.pixel_access(self.readonly)

         # FIXME: future versions should optimize crop/paste
         # sequences!

From fredrik at pythonware.com  Mon Sep 24 18:33:45 2007
From: fredrik at pythonware.com (Fredrik Lundh)
Date: Mon, 24 Sep 2007 18:33:45 +0200
Subject: [Image-SIG] SGI format
In-Reply-To: <46C985DC.F393DA00@wadsworth.org>
References: <46C985DC.F393DA00@wadsworth.org>
Message-ID: <fd8otg$1f2$1@sea.gmane.org>

William Baxter wrote:

> Does anyone have any experience with the SGI RGB format? The PIL
> documentation says SGI images are 'read only'. Image does open them, but
> then it seems you can't get at the data. Any convert, copy, getpixel
> commands give an error:
> 
>     raise IOError("decoder %s not available" % decoder_name)
> IOError: decoder sgi_rle not available

PIL only reads uncompressed SGI images; the reference to "sgi_rle" 
indicates that your images are run-length encoded.

Unfortunately, I cannot say that improving the SGI support is high on
my todo-list; the SGI driver was written in September 1995 (!), and I 
don't think anyone's ever asked about this since then...

</F>


From short.jones.cipher at gmail.com  Tue Sep 25 04:39:04 2007
From: short.jones.cipher at gmail.com (Christopher Jones)
Date: Tue, 25 Sep 2007 12:39:04 +1000
Subject: [Image-SIG] PIL histogram bin 255 empty
Message-ID: <25825ac50709241939wf4fb83dve4ed838805125c9@mail.gmail.com>

With mode I images, the histogram method sometimes returns bin 255 empty,
counting the getextrema maximum value in bin 254 instead. Is this a bug or
is bin 255 treated in a special way, perhaps to count clipped pixels?

Chris Jones
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.python.org/pipermail/image-sig/attachments/20070925/8f9ad06e/attachment.htm 

From fredrik at pythonware.com  Tue Sep 25 09:39:33 2007
From: fredrik at pythonware.com (Fredrik Lundh)
Date: Tue, 25 Sep 2007 09:39:33 +0200
Subject: [Image-SIG] PIL histogram bin 255 empty
In-Reply-To: <25825ac50709241939wf4fb83dve4ed838805125c9@mail.gmail.com>
References: <25825ac50709241939wf4fb83dve4ed838805125c9@mail.gmail.com>
Message-ID: <fdadvs$gne$1@sea.gmane.org>

Christopher Jones wrote:

> With mode I images, the histogram method sometimes returns bin 255 
> empty, counting the getextrema maximum value in bin 254 instead. Is this 
> a bug or is bin 255 treated in a special way, perhaps to count clipped 
> pixels?

what is the getextrema() when this happens?

my guess is that it's a rounding error in the histogram code; it 
basically does

     scale = 255.0F / (max - min);

     for each pixel:
        i = (int) ((pixel-min)*scale);
        if (i >= 0 && i < 256)
            histogram[i]++;

I don't have time to verify this right now, but I guess changing the 
index calculation to:

        i = (int) ((pixel-imin)*scale + 0.5);

would take care of this.

(the whole histogram-of-I/F approach feels a bit odd; not sure why it 
was done that way, really).

</F>


From lucadex at gmail.com  Wed Sep 26 16:30:42 2007
From: lucadex at gmail.com (Luca De Santis)
Date: Wed, 26 Sep 2007 16:30:42 +0200
Subject: [Image-SIG] PIL DPI trouble - Solved
In-Reply-To: <BAY113-DAV15FCB3D9A98A13FAB6984BA3BB0@phx.gbl>
References: <2ce6cd010709210312u7b7d474ct48bb877d7eb0c09b@mail.gmail.com>
	<BAY113-DAV15FCB3D9A98A13FAB6984BA3BB0@phx.gbl>
Message-ID: <2ce6cd010709260730q261998e9q33a946afdda3181e@mail.gmail.com>

> ....
> Any thoughts on this?  Or perhaps can you
> point me to someone who can shed even more light on the matter?

Hi Gary
sorry if it took me some time to get back on this topic.

I made a couple of tests, that I describe below, which seems to
confirm the correct behavior of PIL in managing the dpi of JPEGs. Take
everything "with a grain of salt" since I'm not an expert on the
subject: I just followed an empiric method which seems quite sound.

First of all, I'm dealing with JPEGs so I can't say if what I'm doing
applies to other file formats.
My test environment consists of Python 2.1.3 (I used also Python 2.5
with no apparent changes), PIL 1.1.6 on a Mac OS X 10.4. I'm reading
images with Photoshop 7 plus Preview and Graphic Converter 5.9.5 which
are Mac specific applications.


# First test: reading a JPEG  image #

I started from a simple JPEG created from Photoshop and saved at 150
dpi (image.jpg).
Hexdumping the image I get, as expected, the following values for the
field of the JFIF Segment (cfr.
http://en.wikipedia.org/wiki/JFIF#JFIF_Segment_Format ):
   * Density Unit: 0x01 (dpi)
   * X/Y Density: 0x0096 (150)

I read the image with PIL in the following way:
>>> import Image
>>> i = Image.open('image.jpg')
>>> i.info['dpi']
(150, 150)

Opening the image with the other two applications just gives the same
value for the resolution.

Now I opened again the image from Photoshop and saved it with the
"Save for Web" option, using the name "image2.jpg".
Hexdumping the image I had the following values for the JFIF attributes:
   * Density Unit: 0x00 (No units, aspect ratio only specified)
   * X/Y Density: 0x0064 (100)
The 100 value seems to make sense because, according to the Density
Unit value, it is the ratio (100%) by which you should use/visualize
the image.

Now with PIL you (correctly) can't read the dpi attribute anymore.
>>> i = Image.open('image2.jpg')
>>> i.info
{'adobe_transform': 100, 'jfif_density': (100, 100), 'jfif_version':
(1, 2), 'adobe': 100, 'jfif_unit': 0, 'jfif': 258}

Opening the image again with Photoshop and selecting the "Image size"
option you get 72 dpi but this seems a sort of default behavior of
this application.
Preview for example lefts blank in this case the dpi attribute, while
Graphic Converter must have a bug since it thinks that the resolution
is 100 dpi!

_Conclusion_:
PIL seems to interpret correctly the JFIF attributes. PIL also reads
the EXIF attributes if present: they are stored in the 'app'
dictionary associated to the Image instance.


# Second test: saving a JPEG  image #

Things are a bit different when the JPEG image is saved through PIL,
which _by default_ doesn't handle the dpi attribute (see also below).

In fact, I opened again the 150 dpi image, and saved with name
"image3.jpg" through PIL with the following commands:
>>> i = Image.open('image.jpg')
>>> i.info['dpi']
(150, 150)
>>> i.save('image3.jpg')
>>> i3 = Image.open('image3.jpg')
>>> i3.info
{'jfif_density': (1, 1), 'jfif_unit': 0, 'jfif': 257, 'jfif_version': (1, 1)}

Hexdumping image3.jpg you get:
   * Density Unit: 0x00 (No units, aspect ratio only specified)
   * X/Y Density: 0x0001

Now, I don't really know if the correct value here should be 1 or 100
(I honestly assume 100, so probably this could be a PIL bug).

If you want to maintain the original dpi with PIL when saving the
image just do the following:
    i.save(filename, dpi=i.info['dpi'])

In fact:
>>> i = Image.open('image.jpg')
>>> i.info['dpi']
(150, 150)
>>> i.save('image4.jpg', dpi=i.info['dpi'])
>>> i4 = Image.open('image4.jpg')
>>> i4.info
{'dpi': (150, 150), 'jfif_density': (150, 150), 'jfif_unit': 1,
'jfif': 257, 'jfif_version': (1, 1)}

Hexdumping this new image I got:
   * Density Unit: 0x01 (dpi)
   * X/Y Density: 0x0096 (150)
and all the applications, Photoshop included, recognize the right
resolution (150dpi) for the image.

_Conclusion_:
Just remember to specify the 'dpi' parameter when saving a JPEG whose
resolution should be different than 72dpi.

Sorry if I wrote too much! I honestly don't know other things on that
so I hoped I helped to shed some light on this matter.

Ciao,
Luca

Luca De Santis,
Pisa, Italy

From jefflwise at verizon.net  Thu Sep 27 08:30:53 2007
From: jefflwise at verizon.net (Jeffrey Wise)
Date: Thu, 27 Sep 2007 02:30:53 -0400
Subject: [Image-SIG] An algorithm to find connected groups of pixels
Message-ID: <001801c800cf$f5a76e00$0401a8c0@Willow>

Hi, 

I'm in my first week with PIL and doing well - it's a tremendous library!
I've gotten to the point in my image analysis that I need to find all the
groups of connected pixels in an image.  These groups form arbitrary shapes
in the image.  I want to preserve the shape information for each group, so I
can compare/analyze/categorize... it later.  

The image is composed of pixels of value 0 or 255 in "L" B&W format.  I
assume such an algorithm will need to find a pixel of value 255 and then
check in every direction about it for a neighbor that is 255 too.  The
algorithm would presumably recurse to find neighbors of the neighbor, etc.
I suspect the trick in this algorithm is the data structures needed to keep
track of the discovered adjacencies.  I have loaded the image into the 2D
array-like accessor, so I have free access to each pixel.

I haven't noticed such an algorithm in the PIL yet; maybe I've missed it.
Or perhaps there is Python code available to do this algorithm? Any ideas?

Thanks,
.Jeff Wise



From ggerber at sun.ac.za  Thu Sep 27 11:55:40 2007
From: ggerber at sun.ac.za (George Gerber)
Date: Thu, 27 Sep 2007 11:55:40 +0200
Subject: [Image-SIG] PIL 1.1.5 & 1.1.6 'png decoding error'
Message-ID: <46FB7E1C.3060905@sun.ac.za>

Hi.

PIL 1.1.5 and 1.1.6 cannot load the png:
http://www.ggerber.globesky.com/phd%5Fsite/Stat1Hz_frame_0041.png
I get: "IOError: decoding error when reading image file"

However both versions can load this png:
http://www.ggerber.globesky.com/phd%5Fsite/Stat1Hz_frame_0040.png

Is there a work around for this error?
I would like to work with numpy arrays. Hence using PIL 1.1.6 the code is:

import Image
from numpy import asarray
img = Image.open('C:\Stat1Hz_frame_0041.png')
raw = asarray(img)

Regards
George

From jcupitt at gmail.com  Thu Sep 27 12:45:37 2007
From: jcupitt at gmail.com (jcupitt at gmail.com)
Date: Thu, 27 Sep 2007 11:45:37 +0100
Subject: [Image-SIG] An algorithm to find connected groups of pixels
In-Reply-To: <001801c800cf$f5a76e00$0401a8c0@Willow>
References: <001801c800cf$f5a76e00$0401a8c0@Willow>
Message-ID: <522c6460709270345s6ce29883ke18364ac71d2373c@mail.gmail.com>

Hi Jeff,

On 9/27/07, Jeffrey Wise <jefflwise at verizon.net> wrote:
> I've gotten to the point in my image analysis that I need to find all the
> groups of connected pixels in an image.  These groups form arbitrary shapes
> in the image.  I want to preserve the shape information for each group, so I
> can compare/analyze/categorize... it later.

This is usually done with chain codes. Use a filter (and morphology?)
and a threshold to turn your image into a set of edges, then search
for part of the edge and trace around it, recording up/down/left/right
movements. Once you have the boundary as a set of lines it's easy to
calculate things like area and principal axis.

> check in every direction about it for a neighbor that is 255 too.  The
> algorithm would presumably recurse to find neighbors of the neighbor, etc.
> I suspect the trick in this algorithm is the data structures needed to keep
> track of the discovered adjacencies.  I have loaded the image into the 2D

This sounds very like a flood-fill. You'll find if you recurse for
each connection you will use huge amounts of stack. It's usually
better to keep a list of candidates for later testing and iterate
rather then recurse.

You can also do it a line at a time rather than a point at a time.
Write a function to fill to the right or left until it hits an edge,
then add the point below to your undone list. You need much less
memory and it'll run a lot quicker.

But chain codes are probably a better bet.

John

From janssen at parc.com  Thu Sep 27 18:31:59 2007
From: janssen at parc.com (Bill Janssen)
Date: Thu, 27 Sep 2007 09:31:59 PDT
Subject: [Image-SIG] An algorithm to find connected groups of pixels
In-Reply-To: <001801c800cf$f5a76e00$0401a8c0@Willow> 
References: <001801c800cf$f5a76e00$0401a8c0@Willow>
Message-ID: <07Sep27.093200pdt."57996"@synergy1.parc.xerox.com>

The Leptonica library (www.leptonica.org) has some papers on how to do
this (and code, thought it's in C -- I guess you could call it with ctypes).

Bill

From dblank at brynmawr.edu  Thu Sep 27 19:57:32 2007
From: dblank at brynmawr.edu (Douglas S. Blank)
Date: Thu, 27 Sep 2007 13:57:32 -0400
Subject: [Image-SIG] An algorithm to find connected groups of pixels
In-Reply-To: <001801c800cf$f5a76e00$0401a8c0@Willow>
References: <001801c800cf$f5a76e00$0401a8c0@Willow>
Message-ID: <46FBEF0C.3000908@brynmawr.edu>

Jeffrey Wise wrote:
> Hi, 
> 
> I'm in my first week with PIL and doing well - it's a tremendous library!
> I've gotten to the point in my image analysis that I need to find all the
> groups of connected pixels in an image.  These groups form arbitrary shapes
> in the image.  I want to preserve the shape information for each group, so I
> can compare/analyze/categorize... it later.  
> 
> The image is composed of pixels of value 0 or 255 in "L" B&W format.  I
> assume such an algorithm will need to find a pixel of value 255 and then
> check in every direction about it for a neighbor that is 255 too.  The
> algorithm would presumably recurse to find neighbors of the neighbor, etc.
> I suspect the trick in this algorithm is the data structures needed to keep
> track of the discovered adjacencies.  I have loaded the image into the 2D
> array-like accessor, so I have free access to each pixel.
> 
> I haven't noticed such an algorithm in the PIL yet; maybe I've missed it.
> Or perhaps there is Python code available to do this algorithm? Any ideas?

Jeff,

You can take a look at some code from Pyrobot, Python Robotics, that was 
written by a student:

http://cvs.cs.brynmawr.edu/cgi-bin/viewcvs.cgi/pyrobot/vision/__init__.py?rev=HEAD&content-type=text/vnd.viewcvs-markup

Start with the Blob class. We also have similar code in C++, and wrapped 
by Python using SWIG:

http://cvs.cs.brynmawr.edu/cgi-bin/viewcvs.cgi/pyrobot/vision/cvision/Vision.cpp?rev=HEAD&content-type=text/vnd.viewcvs-markup

If you are interested in image processing, robotics, or other 
camera-based functions, see our course materials here:

http://pyrorobotics.org/?page=PyroModuleComputerVision

-Doug

> Thanks,
> .Jeff Wise
> 
> 
> _______________________________________________
> Image-SIG maillist  -  Image-SIG at python.org
> http://mail.python.org/mailman/listinfo/image-sig
> 


From douglas at paradise.net.nz  Fri Sep 28 01:06:55 2007
From: douglas at paradise.net.nz (Douglas Bagnall)
Date: Fri, 28 Sep 2007 11:06:55 +1200
Subject: [Image-SIG] An algorithm to find connected groups of pixels
In-Reply-To: <001801c800cf$f5a76e00$0401a8c0@Willow>
References: <001801c800cf$f5a76e00$0401a8c0@Willow>
Message-ID: <46FC378F.8090800@paradise.net.nz>

Jeffrey Wise wrote:

> I'm in my first week with PIL and doing well - it's a tremendous library!
> I've gotten to the point in my image analysis that I need to find all the
> groups of connected pixels in an image.  These groups form arbitrary shapes
> in the image.  I want to preserve the shape information for each group, so I
> can compare/analyze/categorize... it later.  
> 
> The image is composed of pixels of value 0 or 255 in "L" B&W format.  I
> assume such an algorithm will need to find a pixel of value 255 and then
> check in every direction about it for a neighbor that is 255 too.  The
> algorithm would presumably recurse to find neighbors of the neighbor, etc.
> I suspect the trick in this algorithm is the data structures needed to keep
> track of the discovered adjacencies.  I have loaded the image into the 2D
> array-like accessor, so I have free access to each pixel.

That's more or less floodfill, as John Cupitt said, and would look
something like this:

http://mail.python.org/pipermail/image-sig/2005-September/003559.html

There's a version of it at the bottom of ImageDraw.py in 1.1.6.

You'll presumably want to change the innermost bit from changing the
colour to shifting the shape onto another image, or something.


douglas