Legacy ImageMagick Discussions Archive

My student Adam Turcotte is finishing a Masters thesis which has, at its core, a quantitative test of upsampling methods based on the common "downsample then upsample back" methodology (flawed as it is). This easily extensible and reconfigurable test suite (if you are willing to write a little python in a *nix environment) will be released as FLOSS. ImageMagick is a key component of the suite.
For this suite, we need a few test images.
If you have one or more digital photograph that you think would be good, please contact me at nicolas.robidoux@gmail.com or through this Forum.
Requirements:

• NO DESTRUCTIVE COMPRESSION (jpeg is out)
• Natural images (no CG or vector graphics).
• Released under a liberal licence (CC-0, CC-by or public domain, for example).
• We'll thank you in the thesis, so you must be OK with your name being associated with the image (and the license).
• Image content should not offend reasonable people of most faiths and political persuasions. If you would not want your children to see it, it's not suitable.
• Raw digital photograph.
• The image should be at least 1682x1682.
• A significant portion should be in sharp focus.
• It should be possible to describe the image as "high quality". If it was not produced with a good digital SLR, it's probably not suitable.

What we'll do with it:

• We'll convert to 16-bit sRGB TIFF and crop to 840x840 (or 841x841).
• We'll remove the metadata (but may copy it in the documentation).
• We'll downsample it various ways, and upsample it various ways, and compute a number of image metrics on the difference.
• We'll distribute it with the suite, and a copy of the cropped version will be shown in the thesis.
• We can't promise that the image won't end up elsewhere (once we put the suite out, it will be up for grabs).

Note: We do not plan on making money from the suite, the thesis, or any associated publication or software (except indirectly: knowledge and a little notoriety).

I have used some images from
http://sipi.usc.edu/database/ (miscellaneous still images)
http://www.hlevkin.com/TestImages/
http://r0k.us/graphics/kodak/ (kodak sample photos)
http://sipi.usc.edu/database/ (arials, time sequence, textures)

The Kodak Sample photo is very good source for real images and specifically for color tests.
You'll have to look at the licencing for each database.

I also noted a list of image databases in
http://www.imageprocessingplace.com/roo ... abases.htm

We're almost done building a short collection of test images.
You can check the current crop here: http://web.cs.laurentian.ca/nrobidoux/misc/test_images/

P.S. Moved to http://web.cs.laurentian.ca/nrobidoux/m ... 840images/.

(The .txt files w/ licenses are not finalized. Please just look. Wait for the final release to spread the images about.)
Note: 840=3*5*7*8=lcm(2,3,4,5,6,7,8).
Comments/criticisms welcome. We are definitely not master photographers/RAW conversion experts.
Note: ideally, these 16-bit uncompressed TIFF images should be imported with an sRGB v4 profile when viewed: http://www.color.org/srgbprofiles.xalter. It does not make an in your face visual difference if you use the usual v2 sRGB, but they really are meant for sRGB v4 w/ Perceptual rendering intent.

The link to your images does not seem to work.

Fred

fmw42 wrote:The link to your images does not seem to work.

Argh! Fixing now. Thank you, Fred.

I'm redoing them all using sRGB v2.

With the kind help of Jean-François Avon, Anthony Barnett, Jana
Duncan, Minglun Gong, Holly Graham, Henry HO, Kirk Martinez, Michael
Muré and Adam Turcotte, I am pretty sure that I have finalized the
images within the first release of the 840x840images CC-by test image
bank. (I still have to finalize the .txt files giving credit,
description, how they were made (some of the descriptions don't match
the latest versions), licensing info, etc.)
It's at http://web.cs.laurentian.ca/nrobidoux/m ... x840images.
Comments/criticism/suggestions welcome.
Note that the images are not compressed because in the context of the
test suite we use them with, we don't want to be decompressing all the
time. Note also that I don't necessarily go for the "best look". I go
for what I believe will provide reasonably "neutral" test images at
the "pixel peep" level, overall.

Have you considered some kind of spiraling fractal? This one is interesting. Oct Spiral_by_FractalMonster.jpg You can search Google for it and find it and some other interesting fractal. I don't know about the licensing. It has, as you know, the same detail at many scales. You can go to http://fractalmonster.deviantart.com/ and see about their licensing. I know you want a tiff not jpg, but you probably can compute something similar of your own, if it fits you criteria. I think GIMP may have some fractal image generator. But if not I am sure there are other fractal generator apps out there.

@Fred: Oyvind Kolas (main dev of GEGL) is of the opinion that resampling methods should be tested on "artificial" images, not photographs.
So, you're not alone in suggesting that I add artificial images to the test image bank, or to the upsampler test suite, or even let the bots take over and forget about "real reality" altogether.
In this incarnation, I won't. It actually would better fit in a different test image bank.
And yes, GEGL (and consequently GIMP) has a fractal generator written by ... Oyvind Kolas (IIRC).

For a image with lots of sharp angles I here by give you permission to use my photo of the
Jewish Holocaust Memorial Berlin, taken by me on 15 April 2006
http://www.imagemagick.org/Usage/img_ph ... t_orig.jpg
This is the original from I downloaded from the digital camera, complete with its EXIF settings.

Here is a small image for preview.

NicolasRobidoux wrote:@Fred: Oyvind Kolas (main dev of GEGL) is of the opinion that resampling methods should be tested on "artificial" images, not photographs.
So, you're not alone in suggesting that I add artificial images to the test image bank, or to the upsampler test suite, or even let the bots take over and forget about "real reality" altogether.
In this incarnation, I won't. It actually would better fit in a different test image bank.
And yes, GEGL (and consequently GIMP) has a fractal generator written by ... Oyvind Kolas (IIRC).

No problem. Just thought you might want to consider that as I did not see anything like that in your suite. But seems you have already given considerable consideration as to what you want to use in this particular test. I particularly like the paint example.

fmw42 wrote:I particularly like the paint example.

Yes makes for a good masking and color changing example! Not that you are using it for that.

Perhaps you should convert the top README text file into a image thumbnail example.
Such as I use in IM example photos.
http://www.imagemagick.org/Usage/img_photos/INDEX.html

Note I classed the rings in this section because they are so large, though really it was generated (though not by me That image is especially good for resize aliasing tests due to visual moire results. But I understand that you want real images.

anthony wrote:For a image with lots of sharp angles I here by give you permission to use my photo of the
Jewish Holocaust Memorial Berlin, taken by me on 15 April 2006

Anthony: Thank you. However:
Two quick notes:

You don't only give it to me if you do: I need the test image bank to only contain images that are CC-0 (Public domain) or CC-by. I'd like the test bank to be widely distributed, so this is not negotiable. (Actually, I was planning to ask you if you would think it to be a good idea for the final version of the test image bank to be somewhere on an ImageMagick server.)

Also: I can't have JPEG artifacts, not matter how small, in images to be included in the test suite. More or less, this means that you need to give the me the original raw image. (There are ways around that...)

(I also think that Adam will have good reason to kill me if I add more images right away.)

(And I just added one: the garland image, courtesy of Mukund Sivaraman, a GIMP dev.)

I understood that it would be public domain. And was permitting this to become public domain for this photo. I made that choice when I put the original camera photo file in IM Examples, from where anyone could copy it!

Law of the Internet: If you can see it or hear it, you can copy it!
Some how... Some way... -- Anthony Thyssen

But I also understand your problem with JPEG. I figured a crop and 3x3 binning (box) reduction would take care of more of those issues. But I understand.

I do not have my camera set to save raw images, I wasn't interested in that high a level of quality. Actually I only just got the camera at that time so I didn't know about raw images anyway -- still don't want them myself