Legacy ImageMagick Discussions Archive

In that case, I'm getting different results: $ convert -version Version: ImageMagick 6.9.0-10 Q32 x86_64 2015-03-14 http://www.imagemagick.org Copyright: Copyright (C) 1999-2015 ImageMagick Studio LLC License: http://www.imagemagick.org/script/license.php Features: DPC HDRI Modules OpenCL Delegates ...

EDIT3: A Q32 build produces the correct result when the output is png64:file.png but an all-black
image when the output file is just q32.png

Do you mean Q64?

I can confirm that the official Gentoo package works fine when set to q32 instead of q64, as well.

Updates: It seems to be due to the configuration flags Gentoo was using. I've built directly from the source and it works. Then I built a second time with this configure line: ./configure --prefix=/usr --build=x86_64-pc-linux-gnu --host=x86_64-pc-linux-gnu --mandir=/usr/share/man --infodir=/usr ...

I've reproduced the bug on the current version of imagemagick, with the same results. $ convert --version Version: ImageMagick 6.9.0-10 Q64 x86_64 2015-03-14 http://www.imagemagick.org Copyright: Copyright (C) 1999-2015 ImageMagick Studio LLC License: http://www.imagemagick.org/script/license.php ...

Example image: https://i.imgur.com/WyBCPty.png Trying to read/convert this with imagemagick fails, eg. $ wget https://i.imgur.com/WyBCPty.png ... $ convert WyBCPty.png test.ppm $ hexdump -C test.ppm 00000000 50 36 0a 38 37 20 38 37 0a 32 35 35 0a 00 00 00 |P6.87 87.255....| 00000010 00 00 00 00 00 ...

How odd; I found that by blurring it *very* slightly (blur factor = 1.01), I got the hash pattern to go away entirely: radius≈3.238 blur=1.01: https://i.imgur.com/YzLMdfk.png ImageMagick seems to replicate this result. It gets even closer with blur=1.015, too. I wonder what the significance of this, ...

Update: I've modified my algorithm to support a fractional radius in an efficient way, so I can actually plug in the true radius of 3.2383154841662362 and see the results. Curiously enough, this still does not remove a hash pattern completely, even when using the “true” formula: jinc(x) * jinc(x * j ...

EDIT: Never mind, I figured it out. Annotations in bold. Hey, I'm trying to implement EWA Lanczos rescaling in my application, and I'm trying to figure out what the best way to do this is. For starters, the underlying problem is that the true radius of the function (at 3 taps) is something like 3 ...