Wednesday, March 2, 2011

A Shark, a Kayak and a Tree Octopus

Back in May 2009 I posted a visual information fluency challenge about a shark and a kayak.

I'm revisiting that challenge because of a new tool I came across recently. Error Level Analysis is a software application that can be used to detect possible jpeg image tampering or touch-ups. To keep it simple, digital images are composed of pixels. Every time an image is copied, the pixels lose some of their original sharpness. The more an image is copied, the more it degrades. Then if something is touched up or added to the copied image, the new part of the image has less error than its surroundings. Read more here.

If you subject the edited image to error level analysis, the newer parts stand out against a background of noise. A good example of this is found at -- roll over the picture of the model on that page.

I ran an information fluency workshop last week at ICE (Illinois Computing Educators) and we used this tool with the image of the shark and the kayak. We also tried it with an image from the infamous Pacific Northwest Tree Octopus site, shown here.

Try it. The original Shark and Kayak picture is attributed to Thomas Peschak. Using error level analysis, the jpeg on his website does not appear to be the original image.  Go to and paste in the url of the image:  See the white outlines of the copyright text? You know that wasn't on the original. The outlines around the shark and kayak are unmistakable.

This poses an interesting investigative situation. Based on all the evidence, except for error level analysis, I'd have to conclude that what this photo shows actually happened. There is so much going for the validity of the story: published in a reputable magazine, a credible author/photographer, even says the image is real.

Then why does it appear to be touched up? And what is the extent of the addition? Maybe the shark and the kayak were merely enhanced. Maybe they were added later. Hard to tell.  One thing's for sure: something is different about this image since it was first snapped.

A second image that HAS to be fake comes from the tree octopus site.  There's no doubt that the tree octopus is a fabrication. Yet when you do error level analysis with, you see what an image that hasn't been photoshopped should look like.  The original photo could have been some type of stuffed or toy octopus intentionally posed on a tree limb--although the octopus appears to have an outline that makes it look out of context.  Blowing up the image might reveal something about that outline.

Clearly, you still have to use your brain and other evidence to tell when an image is real or faked. A tool like error level analysis can help in doing investigation, but you can't rely on it alone.

Challenge: try some favorite images from the web and see what they reveal. Any surprises?

1 comment:

InformationFluency said...

I received an email alert today that Image Error Analysis is closing down. I guess they couldn't find a way to generate any revenue. Too bad, because the software is helpful in investigating "photoshopping." Hopefully, something like this will arise from another source.