In this experiment I invite you to view a sequence of six jpg files that include different versions of the same original photograph.
- All six images seen at the link began as the same original source file – a RAW file from a high quality full frame camera.
- Three versions of this file were created, with each altered in a particular way I won’t yet reveal in advance – let’s call them versions #1, #2, and #3.
- Each of the three versions was then duplicated – in other words two identical copies of version #1 were created, two identical copies of version #2 were created, and two identical copies of version #3 were created.
- The six files were randomly assigned slightly different file names so that in the end we have File A, File B, File C, File D, File E, and File F.
- The six images were then sequenced into a “slide show” version that automatically switches between them, but which also allows the viewer to select them individually.
The Experiment
After you view the linked files in your web browser as presented at the linked page I would like you to consider three questions:
- By viewing the files in your web browser as presented here are you able to perceive any visual differences between the files aside from the letters A-F indicating the different images?
- If you do perceive differences among them, to what would you attribute the visual difference?
- By viewing them in your web browser as presented, can you identify the pairs of identical images?
Click on this link to view the sequence of six presentations of the photograph. The sequence should open in a new window/tab when you click the link. Return here to report your results by adding a comment to this post. (Note: The link was briefly broken – apologies to those who got the missing page error. The problem should now be resolved. Direct URL: https://gdanmitchell.com/images/Experiment2/index.html )
Caveats and Warnings
- I do not make any claims that this is a scientifically constructed experiment – let’s call it an informal investigation of a potentially interesting question.
- If you view the comments of other participants before offering your own response you may be influenced by what they write. Therefore in the spirit of the experiment you should decide on your results before posting your response and before viewing the comment page – and resist the temptation to change your results after reading what others say.
- The whole point of this exercise is to consider what you see when you look at these image in your web browser. If you want to participate, please limit your consideration of the images to viewing them in your browser on the linked page.
- Feel free to look at the images as many times as you want or to switch among them manually. (You might want to mention the extent of your careful viewing in your comment.)
- I’ll stipulate that if you download the files and carefully investigate them that you will possibly find differences whether you can see them or not. But don’t – this isn’t the point here. If you can’t resist doing this please keep your discoveries to yourself until the end. (I’ll also point out that I could do things in the files that might mislead you… ;-)
- You are welcome to hold positive or negative opinions about the aesthetic quality of the photograph, but please keep those out of your response to this experiment.
- I am not saying in advance what the difference are nor am I making any assumptions ahead of time about what the results will be. It is best if you do the same and keep an open mind.
- I plan to reveal everything about the experiment in a day or two.
(This is the second in a series of visual experiments that I invite you to try. You can see the previous Experiment #1 and the followup post about the results.)
After looking at these images on both my NEC Color Proofing LCD (107% aRGB) and my laptop LCD, I was not able to notice the difference. Even when zooming in using the MacOS system level screen zoom feature.
What I see as part of the lesson from this experiment is that below a certain resolution, some noise is not even visible due to the down-rezing process used to create the image. What would be interesting is to re-run this experiment with photos that are 1000+ pixels wide (and maybe not have the fade tradition between photos, that was playing tricks on my eyes at first).
Greg, I sure can’t see a difference at these sizes either. Sometimes, if I stare long enough, I convince myself that I do, but then I find that I most certainly cannot pair up the identical pairs – which I should be able to do if there is a significant difference. I’m sure there is a display size at which the noise starts to become more visible, but I think it is larger than we might imagine.
Dan
I see that you’ve already posted followup, but I’ll add to the chorus.
Pre-read comment:
At the web-size presented, and especially with the fade between shots, I see no difference among the images presented.
Post-read comment:
This appears to be a noise-level experiment that supports the notion that the significance of noise & pixel count are driven by the final image use (size, viewing distance, etc) more than anything else. As a 600-pixel-wide web image, your high levels of noise are imperceptible. At full size (as in your crops above), the difference is clear.
Jeff had a good comment: “how analogous is this type of noise you added to noise generated in the camera? My uneducated suspicion is that the photoshop-generated noise is going to be of a smaller radius than typical camera generated noise and for any given “level” of noise…”
Since sensor noise is generated by the different signal levels of different photosites and my example also varies the levels (and luminosity) of photosites, I think that the effect should be the same or similar. There is one fly in the ointment however – since the camera has to interpolate values from the red, green, and blue sensors, I could see that the noise level variations might actually be less on the sensor since they would be averaged to some extent as a result.
I do think it is safe to say that the levels of noise shown here are very bad. When I look at test results for the 5D2 (one quick example here) you have to crank the ISO up very high to see noise at this level. In any case it is way beyond what you would see at, say, ISO 1600 or ISO 3200. One goal was to produce noise in the samples that would be considered “really bad” if it came directly from the camera.
Another way to think about it is simply to ask: “What would I think if my camera produced noise at this level?” I think we can probably agree that we would regard this as pretty significant noise.
Dan
While I can see that the image changes, there is no meaningful difference between any of them at the resolution posted. The 100% crops show clear differences and this as an excellent demonstration of why viewing at 100% is counter productive in most cases.
Re-looking at the images, I think I can sense just the slightest difference – and my guess as to which is which would probably depend on how long I looked at each (I’d probably change opinions regularly).
But it opens up another question – how analogous is this type of noise you added to noise generated in the camera? My uneducated suspicion is that the photoshop-generated noise is going to be of a smaller radius than typical camera generated noise and for any given “level” of noise (if there was a way to define that so the camera-generated and photoshop-generated could be measured on the same scale), the camera-generated noise would be more visible.
Here are the three source images:
The differences among them are obviously in the amount of noise that was added to the image. No noise was added to the first image – any noise there was in the original capture. 10% level “Uniform” noise was added to the second image in Photoshop. 20% “Uniform” noise was added to the third image in the same way.
The soft photograph was chosen to avoid masking the noise with a lot of other sharp detail – this image provides very smooth gradients from black to white, where noise is typically easier to detect. I also chose this image because it is nearly – but not quite – monochromatic. This meant that I could increase the effect of the noise by using color noise rather than limiting to monochromatic noise – and that the color noise would tend to be more visible against the nearly monochromatic background.
The original set of 6 jpgs includes two from each of the three versions shown here. For last-minute fun you could still try to a) determine whether or not the differences are visible and b) which pairs of images are the least noisy, more noisy, and most noisy.
More information to come later this evening.
Dan
I see no noticeable difference in the images…only in the transition between them.
Decreasing noise: D&F > A&C > B&E
;p
I’m currently looking at it on a small dirty laptop screen, but I see no differences.
No differences.
I see no differences.
OK, I took a longer looked. I focused on the right most bird. I would pair A & C, D & F, and B & E. I would call the A & C pair slightly more pale with less detail. The D & F pair had some slight ghosting artifacts. The B & E didn’t pair as well as the other two, I would call them smooth, but I thought there were slight differences between them. Cheers – Jim
I took a pretty hard look and could not see any differences.
They each have a different letter on them? :-)
I confess to being blind as a bat in this case. At least in the first experiment, I thought I was seeing difference, though after a while I concluded that mostly I was imagining the differences I thought I saw.
I can’t tell any difference.. I don’t even think I can perceive any differences.
Could not see a difference. Could perhaps be a change in DPI and/or file compression, but these eyes could not tell!
I no difference in eyes
I looked at this with a “standard” and a “high def” monitor and see no differences. I even zoomed in to watch the individual pixels. Not sure what this says about compression techniques for web display.
Initially I couldn’t see any difference so I tried zooming in to watch individual pixels, both in the edges of the birds and more into their bodies. Still, I have to say that I see no perceptible difference. I considered switching to a high def display but I don’t think that would make any difference.
These experiments are very interesting… I look forward to the answer. I see no difference.
I would like to think I’m more observant than this, but I see no perceptible difference.
Maybe my browser is screwed up – but the page that loads says content is not available anymore
Oops – bad link! Here is the direct URL: https://gdanmitchell.com/images/Experiment2/index.html
i see no difference
“In eyes?”