Attached Link: Jpeg Quality Perception Test

I recently got into a discussion with a friend about lossy versus lossless image compression and the end result, was, well - a six pack of Guiness' was at stake so things got pretty serious.

Since I'm a big fan of the scientific method, I decided to put together a little test - basically it presents 10 pairs of images. 10 PNGs and 10 JPEGs - and lets you see how well you can tell them apart. I've found that the experiment is quite enlightening. :)

the test is on:
http://www.ranum.com/fun/lens_work/papers/jpegquality/

mjr.

thank you so much for posting this marcus!!! i got 6 out of 10 :] flames? here? oh we're a luvable bunch you know :o)))

I've seen photographers carve eachother up with solarizing filters over this issue. :) I had another fun one come up in the same vein and I mean to find that posting and HTMLize it into a test, too. Someone asserted that "every time you compress/recompress a JPEG the quality goes down" - so, I took a file and compressed it in Irfanview, then Photoshop, then CJPEG/DJPEG, in a loop 10 times and monitored what happened to the file size. Not to anyone's surprise, the file size hovered at about the same point for the entire experiment. I printed the results on my calibrated printer as well as examining the output perceptively. I couldn't tell the images apart nor could any of the dozen or so people I've showed the prints to. The experiment ended when I lost track of which print was which. Maybe we need a "photography myth busters" huh? ;) mjr.

Attached Link: Lossless JPG Rotation

I got 8 out of 10.. But it was real easy with the gradient backgrounds, jpg got a problem with those :)

7/10...Marcus...This is bookmarked and very nicely written. Thanks

i was on my laptop... does that count for my poor grade? :#

Some of the Jpegs were giveways - quality 7,8, etc. I only work at quality=10 or higher. I suspect if I did the test where all the images were q=10 it'd be pretty rare that anyone would perform much better than 50/50/random.

Still - I hope the test gets people thinking! :) Everyone talks about how bad lossy compression is but if you have trouble telling 'em apart in a constructed side-by-side test think how little it must matter to "joe average" when he doesn't have a lossless image to compare against.

mjr.

Message edited on: 07/15/2005 22:24

7:10 I think your conclusions are basically correct; for the final output (either 96ppi for screen or 270ppi for inkjet), one is very hard-pressed to see defects at reasonable levels of compression. I strained, and knew exactly what to look for in the side-by-side images. For archiving of images which will have subsequent processing, non-lossy storage is the best. Enlarging brings out the difficulties most clearly. So does rotation. Even curves. Even cropping! [The 16X16 computation shifts. We have a thread about this somewhere.]

I got 7/10 but I am on a low quality flat LCD. I want to retry this on a CRT. Gotta wait until morning though. I want to forget which is which.

I got 6/10 too! And man, it was hard work!! I burnt my eyes on it!! LOL! Very cool link and test!! Thx for sharing! :)

The third image is really the hard one to tell the dif. Girl in black jacket. Some of the others you can tell by the background which is which.

9:10 on an LCD ........ some were pretty obvious to me..... others not so obvious, I found that the two images on the black backgrounds were the most difficult. By looking at areas along the edges of an area that had light/dark contrast differences, I could see artifacting.

8 o o 10. But it took me nearly a minute per image.

I should maybe do an "extreme" version. Have it start with 10 quality=1 JPEGs against 10 PNGs, then do 10 quality=2... etc. I notice that "8 out of 10" is about typical. I bet the quality=10 and higher images are the tough ones. My next experiment, if I can figure out how to control the variables, is "is camera RAW sharper"? mjr.

OK I am getting a new monitor. I just got 10 out of 10 on my wifes CRT. Take away 1 for memory. I tried not to scroll far enough down to see how I was doing but on one that wasn't possible.

I got 8 out of 10..admitting two were very subjective guesses. PSP8 has a "lossless" rotation in the browser function. File>browser> select image, rt. click.

""is camera RAW sharper"?" While it could be a test, sharpness isn't (necessarily) the reason some prefer RAW. The JPG artifacts don't really show themselves till extreme magnification. But, being able to have more control over the exposure is usually the source of a lot of discussions (smile).

Woo-Hoo 10 out of 10!

Cool test Marcus, i got 10/10 though. Not too difficult spotting the 'stray pix' caused by compression. But i understand you compressed quite a lot. I use psp and rarely go beyond 3% compression, i only use it if theres no other option to get my pics on rr. I use an lcd-screen btw, that must have some impact on my score i think :o) Jeroen