Legacy ImageMagick Discussions Archive

I'm in the process of adding Author/Copyright/Usercomment with exiftool. (A bit late in the game, but better late than never.)

Why no JPEG: The image are used as "gold standards", and results of upsampling from a reduced version are compared with metrics, some of which are very sensitive to (JPEG) noise. Some noise is par for the course, but not JPEG compression noise. (Sure, it's invisible to the naked eye, but not to RMSE.)

Also: I don't want to reward upsamplers that reproduce JPEG noise/blockiness!

A natural use of the test suite (not for the thesis: I mean "future work", most likely for someone else than Adam) is:

Downsample the golden standard images. JPEG compress. Re-enlarge. Compare with the golden standard. What this measures is the ability of the resampler to recover from JPEG degradation.

A variant downsamples a JPEG compressed version of the gold standard, but the final result is still compared to the gold standard.

(Sorry I'm terse. I hope this gives the idea.)

Of course, the reason why I insist on CC-0 or CC-by is exactly so that people can use the images for whatever private tests they want, and that if they republish them somehow they give credit to the nice people who donated them for those who chose CC-by, as well as, hopefully, to those who chose CC-0 and/or public domain.

Believe me, if I had found a copyfree collection of suitably large, high quality well focused uncompressed "natural" 16-bit colour images, I would not have put together my own test image bank. It is very time consuming, in no part because I had to refine my meagre "raw processing" skills. And I'm quite sure that I did not do a perfect job.
The irony of someone who makes a living processing photographs but who himself is not a photographer.

anthony wrote:...
Perhaps you should convert the top README text file into a image thumbnail example...

I'll try to make the distribution a bit more "poking around friendly". Thank you for reminding me to put thumbnails so people can have a quick look see without eating bandwidth like crazy.
But first, let me fix the EXIFs.

Do (some of) the images look too "tame"? I just realized that my processing was done using a Samsung, consequently in your face, uncalibrated monitor (need to buy myself a Spyder...), and I wonder if I went overboard in keeping the colour histograms away from extremes.
For the purpose of computing metrics tame is fine. But I'm wondering if the collection is a bit too tame for human consumption.
http://web.cs.laurentian.ca/nrobidoux/m ... x840images

Anthony (or Cristy): Would you have a good permanent home for the test image bank? Sooner or later, it will need to move out of my former department's server, and it would appear to me that it could be a nice addition to some of the images you guys have floating around. It's at http://web.cs.laurentian.ca/nrobidoux/m ... x840images.

Note that the folder contains a zipped copy of its contents (minus the zipped copy ) and that I produced a folder of 128x128 thumbnails (called "thumbnails") to go along with the folder that contains the actual test images (called "images").

Except for tweaking some bibliography entries (I don't have the info yet), I'm done editing this test image bank. (Man! Is it time consuming )

We can host them @ ImageMagick. They should be available later this evening.

Many thanks Cristy (for this and everything else).

See http://www.imagemagick.org/download/image-bank. Its populating now.

README is still missing. ?

Also a README.html displaying the thumbnails!

What bugs me, is that while the text files contain licence, ownsership, contribution and camera information all you get about the actual photo is one or two words. Where the photo was taken can be of major interest! What footbridge is it? What water tower? Where was that paint peeling? Who is the 'man'.

True this has no real bearing on what the photos are to be used for, but they make the photo more interesting.

anthony wrote:What bugs me, is that while the text files contain licence, ownsership, contribution and camera information all you get about the actual photo is one or two words. Where the photo was taken can be of major interest! What footbridge is it? What water tower? Where was that paint peeling? Who is the 'man'.

This was intentional: I did not want to have to negotiate (and collect) this information from the image contributors, who already had to bear with my demands for time consuming tweaks and "technical" documentation, and I decided to try to respect contributor and subject privacy as much as possible. In other words, this information is missing because I decided to be as "light" as possible on image contributors and subjects.
I also felt that being "opaque" made the photographs stand on their own more. It really should not matter who/what/where/when. Details may actually get in the way of wide adoption.
Look at it another way: Doesn't the "mystery" add a little something?

If contributors or subjects feel like adding details (here, for example), they of course are free to do so. But I don't want to be the vector of this information.