Reader Patrick Lange left the following comment on my “Debunking Claims Made About The Child Victims Of Sandy Hook” entry:
Hey, I am no photoshop expert. Hoaxer claimed that the one with the family standing in front of the christmas tree is photoshopped. I gave him this link in an argument, but he gave me this picture. https://pbs.twimg.com/media/DSwSaV4VQAIRvP6.jpg
He told me “Inconsistent error level. Some areas as bright and even brighter than the shopped on circle. Other areas abnormally dark. Shopped Photo” Can you please explain this?
I’m happy to explain, despite the fact that the clown making these claims couldn’t be bothered. But seeing as how they’re clearly talking out of their ass, I can’t say I really blame them for keeping it so short. They want you to believe that any bright color in an error level analysis means that a photo, as a whole, has been manipulated (beyond the simple addition of the circle), and is therefore fraudulent. That’s not only an overly simplistic of how ELA works, but it’s just not true.
First and foremost, I think it’s important to stress that error level analysis is not something that is taken very seriously by professionals. Actual image forensics expert Jens Kriese has previously described ELA as “subjective and not based entirely on science. This is why there is not a single scientific paper that addresses it” as well as “a method used by hobbyists” which “does not provide clear results”. And Professor Henry Farid has said that it “incorrectly labels altered images as original and incorrectly labels original images as altered with the same likelihood”. So even if this person had interpreted the results of the ELA correctly – and spoiler, they didn’t – well, so what? If people like Jens Kriese or Hany Farid don’t take it seriously, then why shoud we?
But still, what is ELA? How does it work? Or how is it purposted to work? Unfortunately we can’t answer any of these questions without first getting at least a little technical.
In case you’re unaware, JPEGs – which is what we’re working with here – make use of lossy compression to reduce overall file size by discarding (or losing) data. These images will continue to lose data and degrade as they are saved, re-saved, and then re-saved again. What error level analysis does is show the levels of degradation within these lossy images. And since pixels in unmodified images will degrade at the same rate, the end result is a consistent error level. However, in modified images, the altered (or modified) portion(s) will appear bright white, demonstrating a higher error level than the rest of the image. Now we’ve already run into our first problem with this guy’s claim here as there are no bright white areas anywhere in the Avielle Richman photo, which again is indicative of modification. The colors (different from bright white) that we do see are simply the result of high contrast edges. These are especially noticable in areas that face the sun, which is coming in from the right. This is all expected and normal behavior, and that is according to every piece of information regarding error levels and error level analysis that I could find on the Internet. I’ll post these sources at the end of my reply for anyone that is interested… or maybe doesn’t believe me.
Now in images that have been subjected to ELA, the sections that are black correspond to the areas that have not been changed. This indicates a good level of compression with minimal error levels. This exposes another problem with our anonymous friend’s analysis, and this one really demonstrates their ignorance on the matter: there is no such thing as “abnormally dark” when you’re talking about error levels. Again, black – which is about as dark as you can get, I think – indicates minimal error levels. This is exactly what you want to see when you’re looking at an unmodified image as it shows uniform degradation.
So that’s the technical explanation, but there are also a couple of ways in which this claim makes absolutely no sense whatsoever from a logical perspective as well:
- The only part of this photo that would have had to have been edited in order for it to stand as evidence that Avielle Richman was 5-6 years old in December of 2011 would be the cookbook, which shows up as black in the ELA. As we’ve already learned, that means that it is unmodified.
- Given how error levels work, in order for something to be “brighter” than the circle, it would have to have been added after it. And seeing as how this photo is still available on Jeremy Richman’s Facebook page – without the circle that I personally added – this is absurd, if not impossible. If anything was added after the circle, then it wouldn’t exist in the version available on Mr. Richman’s Facebook page. So I’d invite anyone that is still skeptical to compare the two.
Let me further demonstrate with some photos. Feel free to follow along at home.
Here is the original Avielle Richman Christmas photo, as it appears on Jeremy Richman’s publicly available Facebook page:Notice that the cookbook is there and that there is no circle surrounding it. If we download that photo and then run an error level analysis on it using this site, we see the following results:
Mostly black (indicating no modification) and some colors in areas of high contrast. Knowing what we now know about error levels and error level analysis, nothing here stands out as suspicious in the slightest.
Now we’ll take my version of this photo – which has the circle around the cookbook – and run an error level analysis on it using the same site:
Finally, let’s compare the results of these two photos:
Outside of the circle and additional compression artifacts – the result of further image degradation, wholly expected with subsequent saves – there is no difference. There are no bright white areas indicative of tampering. Outside of the circle in the second/right photo, which I obviously and admittedly added myself, there is no evidence of “Photoshopping” here. The claim is total nonsense.
One last demonstration, and I think this one may be the most useful:
Using a stock photo, I “Photoshopped” (or added) a cat onto the dog bed in the Richman Christmas photo, as seen here:
Not too bad, right? Then I ran an error level analysis on it, using the same online tool I used earlier. Here are the results:
The cat very clearly stands out and is the bright white color we should expect now that we actually know how error levels and error level analysis works. Compare these results to the original photo as well as the photo in which the circle has been added and the differences are crystal clear. The Christmas photo has not been “Photoshopped” in the sense that any information was added to it.