What's new
  • Happy Birthday ICMag! Been 20 years since Gypsy Nirvana created the forum! We are celebrating with a 4/20 Giveaway and by launching a new Patreon tier called "420club". You can read more here.
  • Important notice: ICMag's T.O.U. has been updated. Please review it here. For your convenience, it is also available in the main forum menu, under 'Quick Links"!

Detecting photo modifications (Photoshopped etc) made easy with Error Level Analysis

PhenoMenal

PhenoMenal

Hairdresser
Veteran
I came across this today, pretty interesting and surprisingly EASY way to detect modifications in photos... it's called Error Level Analysis (ELA), and there's a website that allows you to upload photos for free and get an instant analysis image - so you don't have to understand it or be a techy

To generate an ELA image the website is http://fotoforensics.com which allows you to upload an image, or give it a URL of an image, and they also have a Firefox plugin so you can simply right-click on any website image to have a popup-menu option. Their only stated rule is no porn of any description (not that you need to check that anyway -- something is gonna be enlarged), so I'm assuming cannabis-related photos are fine. The site has a FAQ, and many interesting tutorials etc about not just ELA but also a few other photo forensic tricks, such as EXIF metadata in which some smartphones/cameras record GPS co-ordinates, so it's an interesting read.

ELA Background:
In August 2007, Dr. Neal Krawetz gave a presentation at the Black Hat Briefings computer security conference. The presentation, titled "A Picture's Worth", covered a handful of novel photo analysis algorithms. (YOUTUBE of the presentation: Part1, 2, 3, 4). Using these algorithms, researchers can determine if a picture is real or computer graphics, if it was modified, and even how it was modified. Dr. Krawetz gave variations of this presentation at different conferences between 2007 and 2010.

Following the disclosure of these algorithms, many people began recreating them. Error Level Analysis (ELA) is one of the simpler algorithms, and many people implemented their own variants. In 2010, Pete Ringwood created the "errorlevelanalysis.com" web site as a free service where people could submit photos and web pictures for analysis. The result was an instant hit.

In 2012, Mr. Ringwood decided to retire the site, which had introduced millions of people to the field of photo forensics. Hacker Factor has recreated the service as "fotoforensics.com", maintaining the basic principles that Pete Ringwood established: a free service that provides an introduction to photo forensics.
Click to expand...

Here's a recent viral photo doing the rounds at the moment, the original has him holding a passport or something, but somebody Photoshopped™ a floppy disk over it ... it's a simple one but visually quite effective in regards to maintaining photo-realism:
analysis.php


Here's the original, unedited version:
analysis.php


Enter ELA ...
Error Level Analysis (ELA) is an algorithm that evaluates the error level potential of a JPEG image. JPEG is a lossy image format; every resave degrades the picture. The amount of degradation varies based on the number of saves. The first save loses a lot, the second save loses a little more, and by the 20th save, it is probably as low quality as it will ever get. When a picture is modified, the changed parts have a higher error level potential than the rest of the image. ELA works by saving the picture at a known quality level (like a JPEG at 95%), and then determines how much changed. Edits and splices appear as regions with more change. ELA Tutorial
Click to expand...

When we use ELA to look at the original unedited photo we don't see anything unusual, nothing stands out:
analysis.php


But despite the Photoshopped version looking so convincing, when we look at the ELA version we get a white glowing disk of white noise that screams out "ive been Photoshopped!":
analysis.php


IMPORTANT: Sometimes you see white-noise in untouched images, but you'll find it's consistent. We're not actually looking for white-noise (some cases the edit will show as color-noise when everything else appears as white-noise, there's an example of that on page 2 of this thread) -- what we're looking for is INDESCREPENCIES/INCONSISTENCIES - as you can see in the ELA pic everything looks consistent except for the floppy disk.
 
Last edited:
PhenoMenal

PhenoMenal

Hairdresser
Veteran
Another example ... here is the original, unedited image:
books-resave.jpg


Now lets look at a Photoshopped version ... can you tell what modifications have been made? If you compare them closely you'll see books have been _duplicated_ from the original image and a dinosaur _added_ from a completely separate image, demonstrating two types of modification ... but it's not easy to tell, especially the books:
books-edited.jpg


... until you look at the ELA version below, then it becomes clear exactly what has been modified (the white noise sections) - including exactly which books, even though the duplicated books essentially look the same in the previous Photoshopped image:
books-edited-ela.png


A COUPLE QUICK NOTES

- In case you're wondering, ELA only needs 1 photo, the so-called "after" photo ... it doesn't use or need a "before" photo to make comparisons - it isn't a comparison-based test in that sense, making it very useful for detecting image modifications when you don't have access to the original source image.

- The ELA technique actually has nothing specifically to do with Adobe Photoshop (i only used the word "Photoshopped" as an example because it has joined our vernacular in the same way "Googled" has). ELA basically applies to all image editors (Photoshop, GIMP, Microsoft Paint etc etc) when image modifications are made to lossy image formats such as JPEG and PNG.

- Having said that, ELA analysis _can_ hint at which program was used to make the image, and ironically Photoshop is the easiest to detect - for example, by detection of 'rainbowing'. Also, if Photoshop's "File > Save For Web" option was used to save the image then even the Quality % setting can be determined often with a high level of accuracy.
 
Last edited:
HUGE

HUGE

Active member
Veteran
Someone hurry and check out the sandy hook photos that everone was saying were phototshopped.
 
PhenoMenal

PhenoMenal

Hairdresser
Veteran
Poopy,
No, the ELA version of the image might seem a bit misleading in that case if you're new to ELA but if you compare it to the original (you basically need to OVERLAY them and flick back and forth ... and by 'the original' I don't mean you need access to the original image, I just mean the un-ELA'd version) you'll find they're actually perfectly consistent. Likewise, nothing in the ELA actually 'stands out', and that's generally the giveaway -- artifacts that stand out.

Sometimes you can see a modification really easily, such as in the previous examples in this thread, but othertimes you'll need to flick back-and-forth between the ELA and original non-ELA images. (the fotoforensics website provides this capability by default when you upload there, just drag your mouse cursor on-and-off the image)

For example, there is a lot of white-noise in your Obama ELA, but it's all very consistant across the image, and none of it stands out ... either the _ENTIRE_ birth certificate was Photoshopped _perfectly_, or it's real. Not to mention it doesnt make any sense to modify _every single field_ of a birth certificate (why not just make one from scratch)

These are some of the little intricacies that people 'reading' ELA images need to know about before making a final judgement :)
 
Last edited:
HUGE

HUGE

Active member
Veteran
The one with the girl in the red dress sitting on her dads lap and any of the "crime scene photos"
 
HUGE

HUGE

Active member
Veteran
I'm at work on a phone. Was hoping someone sitting in front of a pc could do it to ease my curiosity while I'm working ng
 
PhenoMenal

PhenoMenal

Hairdresser
Veteran
Also keep in mind that when using ELA you want to find the original, highest-res version of the image ... if a news agency has taken it and resaved it then some of the ELA aspects may be watered down making it trickier to detect (but again, ELA vs non-ELA overlap flicking back'n'forth can often make it easy here)
Take some time out to read the ELA tutorials, FAQs etc, it's quite an empowering technique! :)
(and makes the "forensics technique" of "zooming in" look elementary in every way) :)
 
Last edited:
medmaker420

medmaker420

The Aardvarks LED Grow Show
Veteran
PhenoMenal

PhenoMenal

Hairdresser
Veteran
i just did one of one of my led bud shots
i wanted to see how it detected the led grow lights
Click to expand...
med you can see the uniformity of the ELA version :)
The beauty of ELA is that modifications usually stand out like a sore thumb
 
medmaker420

medmaker420

The Aardvarks LED Grow Show
Veteran
PhenoMenal said:
med you can see the uniformity of the ELA version :)
The beauty of ELA is that modifications usually stand out like a sore thumb
Click to expand...
I thought it was pretty cool because with my camera and video camera they don't have that white balance deal so my videos and pics usually are pink,purple,blue,red when taken so I wondered if that naturally different light during the picture taking affected this at all.

I found the brighter green on the hairs interesting because even from the original picture they look bright due to trying to overtake the led spectrum in the tent.

really cool stuff with leds actually BUT aside from that

this is fun to do with regular photos, I can see this being a blast to do with online "models" photos etc to see how modified those pics are :biggrin:
 
PhenoMenal

PhenoMenal

Hairdresser
Veteran
medmaker420 said:
this is fun to do with regular photos, I can see this being a blast to do with online "models" photos etc to see how modified those pics are :biggrin:
Click to expand...

This chick didnt have any lipstick, so i selected her lips and adjusted them with Hue & Saturation (some of my finest work), so i didnt actually add to or alter the image per se, just a minor subtle adjustment along the same vein as adjusting brightness or contrast ...
TEST.jpg


Still sticks out like a sore thumb though:
analysis.php


In my first post in this thread the Kim Jong-Un ELA modification (the floppy disk) sticks out as white-noise, whereas in this example everything else is white-noise whereas the edit is color-noise ... good example that we're not looking for white-noise, but INDISCREPANCIES/INCONSISTENCIES, often sticking out like a sore thumb :)

http://fotoforensics.com/analysis.php?id=daece695a5f6bc97e06dc8561eea79457560faff.17053
 
Last edited:
You must log in or register to reply here.

Latest posts

Latest posts

Top