|  support |  documentation |  report a bug |  advanced search |  search howto |  statistics |  random bug |  login
Bug #39273 imagecopyresized() not compatable with alpha channel
Submitted: 2006-10-27 04:17 UTC Modified: 2006-11-04 02:26 UTC
From: seth at pricepages dot org Assigned: pajoye
Status: Closed Package: GD related
PHP Version: 5.1.6 OS: Mac 10.4
Private report: No CVE-ID:
 [2006-10-27 04:17 UTC] seth at pricepages dot org
imagecopyresampled() should be copying the alpha channel, but 
it doesn't seem to be doing so. This is a palette based source 
image being copied to a true color image.

If you use imagecopy() instead, the image copies as expected 

Reproduce code:
$small = imagecreatefrompng('');

$width = 300;
$height = 300;
$srcW = imagesx($small);
$srcH = imagesy($small);

$img = imagecreatetruecolor($width, $height);

//Make a transparent canvas
$trans = imagecolorresolve($img,255,255,255);
imagecolortransparent($img, $trans);
imagealphablending($img, false);
imagefilledrectangle(	$img,
						0, 0,
						$width, $height,

//This shouldn't *need* to be on, but it does
imagealphablending($img, true);

//One of these works, the other doesn't
//imagecopy($img, $small, 0,0, 0,0, $srcW, $srcH);
imagecopyresized($img, $small, 0,0, 0,0, $width, $height, $srcW, $srcH);

header('Content-Type: image/png');

Expected result:
An enlarged, pixellated, mostly transparent, image.

Actual result:
A black, opaque, image.


Add a Patch

Pull Requests

Add a Pull Request


AllCommentsChangesGit/SVN commitsRelated reports
 [2006-10-27 07:42 UTC]
Please try using this CVS snapshot:
For Windows:

 [2006-10-27 14:01 UTC]
There is nothing wrong in imagecopyresize (or imagecopy).

The problem you have is the misunderstanding of what is the background color, the alpha channel and alpha blending.

Your original image has many black colors, one is transparent (what you consider as background), and the other with various transparency levels.
Try the code below, it will explain you what is your image and how it works.
$small = imagecreatefrompng('39273.png');
$width = 300;
$height = 300;
$srcW = imagesx($small);
$srcH = imagesy($small);

print_r(imagecolorsforindex($small, imagecolorat($small, 0,0)));

// one of the ""black"" pixel
print_r(imagecolorsforindex($small, imagecolorat($small, 37,1)));
imagecolortransparent($small, 1);

$img = imagecreatetruecolor($width, $height);
$trans = imagecolorresolve($img,255,255,255);
imagefill($img, 0,0, $trans);
imagecolortransparent($img, $trans);

imagecopyresized($img, $small, 0,0, 0,0, $width, $height, $srcW, $srcH);
imagepng($img, '1.png');

 [2006-10-27 17:22 UTC] seth at pricepages dot org
I am aware that the image is fully black, except for 
variations in alpha. That is why this is a bug related to 
the alpha channel and not any other.

Well, I went in and fixed it myself. The problem was in the 
function gdImageGetTrueColorPixel(). It assumed that palette 
images always have binary transparency, with a correct value 
in im->transparent. Because my source image didn't have a 
correct value in im->transparent, it was always marked as 

This line:

return gdTrueColorAlpha(im->red[p], im->green[p], im->blue
[p], (im->transparent == p) ? gdAlphaTransparent : 

Needs to be changed to:

return gdTrueColorAlpha(im->red[p], im->green[p], im->blue
[p], (im->transparent == p) ? gdAlphaTransparent : im->alpha

Making this patch also fixes the same bug in 
imagecopyresampled(), and who knows what else.

Although, I would recommend using gdTrueColorAlpha() at the 
appropriate point(s) in gdImageCopyResized() instead of 
gdImageGetTrueColorPixel(). This would eliminate an extra 
function call, branch, and color lookup.
 [2006-10-28 10:49 UTC]
"I am aware that the image is fully black, except for 
variations in alpha. That is why this is a bug related to 
the alpha channel and not any other."

You are not aware that your image is *NOT* fully black.

Background color (the transparent color) is not the same than a color with a ALPHA component. Please run the code I gave you, read the results (like the values of the channels in these two pixels, or any other).

 [2006-10-28 13:15 UTC] seth at pricepages dot org
I ran your code before I posted my last comment, but I still 
don't know what you are implying. The results are exactly as I 
expect them to be.

If this bug is bogus, please tell me how I can create an 
enlarged version of test.png? It isn't possible without 
applying a fix.
 [2006-10-28 14:04 UTC]
Use the code I just gave you, it does create the resized version (by copying):

 [2006-10-28 14:11 UTC] seth at pricepages dot org
But that has one of the bugs that I pointed out: boolean 
transparency. The original doesn't have that problem.
 [2006-10-28 14:19 UTC]
Pardon? What is a boolean transparency?

Please understand these three things:
- The transparent color is one *color*, an index for palette  based image or 32bits value for truecolor images. It defines which color is used as the *background* color (like white is   the background color of a white paper).

- The *alpha* channel of a pixel defines how translucide the pixel has to be. It has nothing to do with the transparent color.

- Your image has *no* transparent color but many pixels with various *alpha* levels.

Load the result images in your favourite paint programs to see what I mean. The area outside the line is translucide, it is due to the alpha channel.

 [2006-10-28 14:43 UTC] seth at pricepages dot org
translucide (French) = translucent (English)

Sorry, that confused me for a bit.

"Boolean transparency" means that the pixel is either fully 
transparent, or fully opaque. Never partially transparent.

The area outside of the line is fine, it should be 
transparent and it is in your image. This is correct.

What is suffering from "boolean transparency" are the edges 
of the line. For example:

In your example script, one alpha value was 63, and the 
other was 127. In your output PNG, those values have been 
changed to 0 and 127.

I should mention that if you are using Microsoft's Internet 
Explorer version 6 or less, a PNG image will display with 
boolean transparency due to a bug in the browser. So you 
won't be able to see the difference. Use FireFox.
 [2006-10-28 15:14 UTC]
Sorry for my bad wording (images "speak" better), but we are getting somewhere, finally. We were talking of two different problems. I was talking about a common misunderstanding and you about a more specific problem. Thanks to have insisted and nice catch :)

Check this image:

Is it what you expect?

 [2006-10-28 15:26 UTC] seth at pricepages dot org
I'm American; if I'm exposed to more than one language my head 
explodes. ;)

Yep, that image is exactly what I'm looking for.

Thanks for not just ignoring me after marking the bug bogus, 
I've had that happen also...
 [2006-11-04 02:26 UTC]
It will be available in 5.2.1 or later. Thanks for the reproduce script.
PHP Copyright © 2001-2017 The PHP Group
All rights reserved.
Last updated: Wed Jan 18 05:14:51 2017 UTC