Perceptual Image Diff

[Freddie Witherden]

Hi all,

While GSoC is now (sadly) over, it will take more than a date in the calendar
to get rid of me.

While looking through the cairo-dev mailing list today I saw a question
regarding output consistency. In the thread it was mentioned that the Cairo
test suite first attempts to perform an exact pixel comparison and if that
fails will compare the generated image to the reference data using pdiff.

So, I looked up pdiff and found it to be: http://pdiff.sourceforge.net/ which is
a tool to compare perceptual differences between images.

As bitmap comparisons are also a problem for mathtex I think it is worth
implementing.

My current plan (assuming everyone likes the sound of it) is to:
- Add the reference ,png images from mathtex to svn;
- in the utility check to see if pdiff is available;
- change the utility to generate different file names when running the tests;
- call pdiff to compare the two.

Regards, Freddie.

[Michael Droettboom]

Seems very promising.  I'll be paying special attention to this as it could be very useful for matplotlib as well.

Mike

--
Michael Droettboom
http://www.droettboom.com/

Sent from my MacBook Wheel

This message is confidential, intended only for the named recipient(s) and may contain information that is privileged or exempt from disclosure under applicable law. If you are not the intended recipient(s), you are notified that any dissemination, distribution, or copying of this message is strictly prohibited.

[Freddie Witherden]

Hi,

In order to evaluate PerceptualDiff I need some failed tests! If anyone could
tarball me any and all tests which fail the bitmap comparison (as indicted by
the test utility) I would be more than grateful.

Regards, Freddie.

[Michael Droettboom]

Sure.  Here's a tarball of my tests directory.

(Note, I found a buglet that stixsans fonts weren't being tested at 300 dpi and fixed in the SVN -- you may need to pull that to compare against my included images).

Mike