Photoshop and fishing don’t go hand in hand
How to Spot a Photoshopped Image
BountyFishing works like so:
- Participants go fishing and photograph what they catch
- Participants then submit these photos to the BountyFishing website
- BountyFishing awards cash and prizes for the longest validated catch
The biggest hurdle we faced was how we were going to validate and authenticate the fish photos our users submitted. Photos altered with Photoshop can be incredibly convincing. With a few simple clicks of the “Quick Selection Tool” the fish shown below (top left) was selected, dropped into a separate layer, and stretched to add more than an inch to its length. With a little more work to move the shadow and soften any rough edges, the resulting image (top right) shows no signs of tampering.
Unaltered image
Photoshopped image
The solution we found was to use a piece of software developed by Hany Farid, a professor at Dartmouth College who is renowned in digital forensics. When the fish was stretched, Photoshop filled in the missing pixels by interpolating their values from the original recorded pixels. These regularly spaced new pixels are a specific combination of their surrounding pixels. Such regularities rarely occur in natural images, so their presence can be used as evidence of tampering.
Shown below is the output of the BountyFishing software that detects these correlations – the red-coded regions corresponds to the doctored portion of the image.
This software is used by federal law enforcement agencies and can detect various forms of tampering.
We have exclusive rights to the software for measuring fish, but if you really feel like getting your hands dirty with digital forensics Hany Farid’s website has an interface to MatLab for manipulating and analyzing digital images. In addition to the software above, a few other dead giveaways that a photo has been altered are:
- Excessive cloning - repetition of a particular piece of the photo, often used to stretch or morph images.
- Inconsistencies in lighting and noise. The easiest way to spot altered lighting is to increase the contrast of a photo so all the differences in lighting are exaggerated.
- Look for shadows that don’t match up what’s casting them.
- Optical aberrations, including patterns that aren’t seamless or appear to be inconsistent (or artificially consistent) with the surrounding area.
- Using human anatomy and other reference points BountyFishing verifies the authenticity of the ruler’s length.
We’ve got a flash demo of how we authenticate photos that gives a little bit more information about how this all works.
i doubt forging would
trouttotinpete said this on June 19th, 2007 at 7:31 am
Wow, really interesting stuff
Greztky likes to Fish said this on June 19th, 2007 at 7:42 am
They should use that software everywhere, there is way too much bullshit going on, on the internet.
Ankur said this on June 19th, 2007 at 9:29 am
Wow! Something like this is long time due.
E Pluribus Unum said this on June 19th, 2007 at 10:13 am
Workaround: Use a scaled-down replica ruler that is the exact same relative dimensions as the regulation ruler.
Leapin_trout said this on June 19th, 2007 at 10:23 am
Hey, I thought that too but if you read about how they validate:
“Using the human eye and other reference points in the photo of you holding your catch, the software verifies the accuracy of your ruler”.
So they seem to have this covered.
VRH said this on June 19th, 2007 at 11:05 am
Hey! That’s a Northern Longnose Bass! I thought they were all but gone in the US.
Bob said this on June 19th, 2007 at 11:17 am
That was the last one.
RIP Northern Longnose Bass (4000BCE - 2007CE)
John said this on June 19th, 2007 at 11:21 am
Well they are all gone now…that was the last one.
Lisa said this on June 19th, 2007 at 11:36 am
Photoshop images are getting professional nowadays. It is pretty difficult to identify professional images.
Lisa
free disposable email addresses
http://www.spaml.com
Keer said this on June 19th, 2007 at 11:37 am
Northern Longnose Bass = pwn3d
Tony said this on June 19th, 2007 at 11:42 am
It was delicious!
Fishing TV said this on June 19th, 2007 at 11:42 am
Quite interesting, do a lot of people try to pass of photoshopped images?
roger said this on June 19th, 2007 at 12:05 pm
What you do is make an authentic looking measure that is scaled down to, say, 75% of an actual tape. Then you measure your fish, take the picture. No Photoshop required.
Hardcore said this on June 19th, 2007 at 12:37 pm
Read the end of the article “Using human anatomy and other reference points BountyFishing verifies the authenticity of the ruler’s length”. They do this with the image of the person holding the fish + you are required to complete an affidavit…
Go-Referencement.org » Les réseaux sociaux s’en vont à la pêche said this on June 19th, 2007 at 1:07 pm
[…] 000 $US en jeu cet été ! Mais comment valider les poissons pêchés en ligne ? D’après un article publié sur le blog de BountyFishing, le système de validation des photos est fourni par Digital […]
Dell said this on June 19th, 2007 at 1:10 pm
Very interesting article and an equally amazing piece of software. Unfortunately, i know photoshoppers the world over will find ways and means of preventing detection of their photoshopped images.
John Lehmann said this on June 19th, 2007 at 1:23 pm
What prevents:
1. Using a picture of someone else?
2. Using a picture taken on a different day?
They could verify #1 with other pictures? or ID I guess (they don’t mention this, but what about #2?)
Leapin_trout said this on June 19th, 2007 at 2:01 pm
From what I saw in the flash demo you need to have the code for the day you are fishing on viewable in the photo. I would assume that you would of had to purchase a pass for that day to be able to enter the photo.
Scott Schomburg said this on June 19th, 2007 at 2:11 pm
There is a couple things wrong with this example, the machine finds stretched pixels so instead of stretching the fish couldn’t you just enlarge (horizontally AND vertically) the fish equally therefore eliminating any stretched pixels? Another handy trick would be to alter your photo using the clone tool (Photoshop) which doesn’t alter pixels it simply copies unaltered pixels already on the image. The biggest problem here is that anytime you resize your image (say to create a jpeg for a website from a highres photo) the computer is interpolating the pixels, so all you would have to do to cover your tracks is make your alterations on the highres version and then just alter it’s size a couple times making all the pixels even.
And one more thing “Using human anatomy and other reference points BountyFishing verifies the authenticity of the ruler’s length” is a lie. Using human anatomy from personal snapshots to verify a rulers accuracy is impossible… completley impossible. And what are these “other reference points” they use?
I find the article interesting but most of it seems to greatly exaggerate the current capabilities of computers
I am now offering my services (for a small fee) for anyone who wants there fish enlarged in Photoshop. It comes with my personal gaurantee that they won’t be able to detect an alteration. Email me scott@scottschomburg.com if you would like to win some fishing awards.
Eric said this on June 19th, 2007 at 3:50 pm
Fantastic info on digital tampering. I do a lot of digital artwork and I am always amazed at the stuff people can (and do) do with Photoshop.
Eric
Michael said this on June 19th, 2007 at 4:55 pm
and Scott Schumburg has just proved himself to be someone who knows nothing about modern detection software….
asdf said this on June 19th, 2007 at 6:41 pm
Scott Schumburg also proved that he doesn’t know English by saying “there” instead of “their” in the last paragraph.
jason said this on June 19th, 2007 at 9:51 pm
i’m going with scott on this one…maybe not a full 100%, but they cant use the human eye as a reference point to a unit of measure. Simply because everyone’s eyes are different, both in shape and size. One person can have small eyes and big hands, and another have big eyes but small hands. what about midgets? (no pun intended) I mean seriously that part of the article is a bit ridiculous. same goes for the fags that worry about grammar on a freakin comment board.
Shinda said this on June 19th, 2007 at 10:15 pm
Or you could just open the image in notepad and look at the first line which contained the Photoshop tag ;), but for really clearly some people who feel they can circumvent the software didn’t read the article. It reads the image finds photoshop patterns and then highlights those patterns. Saying you can still Photoshop these images without being caught is saying that you’re not using photoshop and instead have another program which uses a completely different algoritham and pattern for image manipulation.
Nick said this on June 19th, 2007 at 10:44 pm
Scott,
I am fairly familiar with Prof. Farid’s algorithm and it does detect cloned pixel patterns in addition to the interpolated pixels mentioned in the article. Simply using the clone tool would trigger it. These patterns would largely remain intact (at least enough to trigger it) though resizing and compression.
As for resizing on 2 detentions; you will still have the problem of inconsistent interpolation between the fish and the background since the one is being interpolated and the other not. Resize the picture all you wish the differences will still be apparent when you pull the image apart and pass it though various algorithms. Will it catch everything? Of course not, some of the oldest out of camera tricks such as forced perspective are unlikely to trigger it. That said the software he has developed is amazingly robust and I encourage you to actually try it before passing judgment on what it can and cannot do. You will likely be surprised.
Scott Schomburg said this on June 19th, 2007 at 11:40 pm
I may not know the difference between “there” and “their” people but I went ahead and included my full name along with my opinion (rare on the internet) AND I do know that it is spelled S-C-H-O-M-B-U-R-G, I deserve that at least. The time in art school that I wasn’t learning english I spent in front of a computer using photoshop and for all the people disagreeing I don’t see any explanations for why I “know nothing about modern detection software”. For the contest they want your image to be between 2200 and 3264 pixels wide, so most editing (done by someone experienced) at 5000 pixels and up and then shrunk to 2500 pixels would be undetectable (I am guessing) When going from a larger image to a smaller image Photoshop is redrawing the entire picture, I propose that any editing that takes place on the larger version would be lost in this shrinking. I think the thing that convinces me the most that the software has it’s inadequacies is the fact that they list 5 ways that they test for forgeries (not including their super computer) - if their computer was omnipotent they wouldn’t need to mention they “Look for shadows that don’t match up what’s casting them.”
I am not planning on now making my fortune off of fake fish photos and I hope that everyone in this contest plays by the rules and all that but I love photoshop and related discussion. I am going to try and get ahold of the bountyfishing people and see if I can put there machine to the test with this 8ft. Coho Salmon that I caught in the pond behind my house…So I will look into it more and if Prof. Farid’s algorithm is all it’s advertised I will photoshop a picture of myself eating crow! A big crow!
And one last thing, I think removing the photoshop tags and converting it to an image file as if it had just come off my camera would be the easiest part of the whole scam.
Sekularista » Blog Archive » Pictures are not to be trusted said this on June 20th, 2007 at 3:26 am
[…] the tools become more and more sophisticated. Take a look at the two pictures below, taken from a website dedicated to fishing. The first one is the original, unaltered picture; the second one is a picture with about an […]
corey graham design » The Magic CSI Image Enhancement said this on June 20th, 2007 at 6:25 am
[…] the photo to make the fish bigger, and how they’ve taken measures to spot doctored photos. Read the article here, it’s pretty […]
Bounty Admin said this on June 20th, 2007 at 6:46 am
With regards to Scott’s post
Scott wrote:
> There is a couple things wrong with this example, the machine finds
> stretched pixels so instead of stretching the fish couldn’t you just
> enlarge (horizontally AND vertically) the fish equally therefore
> eliminating any stretched pixels?
No. We are not detecting “stretched pixels”. We are detecting
interpolation that results from making something bigger (or even
rotating). Whether you enlarge in one direction or both directions,
this manipulation leaves gaps in the image that have to be
interpolated. Read more about this technique here:
http://www.cs.dartmouth.edu/farid/publications/sp05.html
Scott wrote:
> Another handy trick would be to alter your photo using the clone
> tool (Photoshop)
As we described in the article, we have a tool that can detect this
form of tampering. Read more about this technique here:
http://www.cs.dartmouth.edu/farid/publications/tr04.html
Scott wrote:
> so all you would have to do to cover your tracks is make your
> alterations on the highres version and then just alter it’s size a
> couple times making all the pixels even.
This wouldn’t work because you would then submit an image that is not
of the required native resolution. And you will have introduced at
least three other artifacts in the image that we can detect.
The Bounty Admin
Christopher Dack said this on June 20th, 2007 at 7:14 am
Well, the first mistake is going for the obvious — enlarging the fish and leaving everything else the same. This is what causes Photoshop (or other image editor) to add telltale pixels that weren’t there in the original. It’s virtually the same amount of work to shrink everything except the fish, accomplishing the same result but going at it from the opposite direction. Instead of adding pixels, Photoshp throws away pixels in the portion being shrunk, which is MUCH harder to detect. If you’re careful with the areas where the (shrunken) hand in the photo interacts with the fish, you can make it much tougher for experts to discover the fakery.
Kula bácsi said this on June 20th, 2007 at 8:21 am
Scott Scumburg is a moron.
Michael Verry said this on June 20th, 2007 at 8:37 am
There is an easier solution, have all digital images submitted in raw format. Then any alterations will be found in the properties of the image.
Jonah said this on June 20th, 2007 at 8:57 am
Scott Scumburg IS a moron.
Hey idiots, did it occour to anyone that you can just photograph a scaled down measuring tape? Duh! Scale in a photograph really only pertains to familiar objects of known sizes. (people, currency, etc.) I would rather just print out a tape that is 1/2 scale to make my fish look twice as large.
Scott Schomburg said this on June 20th, 2007 at 9:49 am
Ok so after reading through all the rules it turns out that you are not allowed to mess with the size of your image once you have taken it off the camera. Which makes it much harder to fake because then you can’t hide the manipulation within a higher res version because they allow no manipulation whatsoever. Apparently there is some RGB pattern created on the image by the light sensor on the camera that the software can detect, and change in that is what the program finds. That might be a little bit harder to work around but I am guessing after manipulation you could load the image back on the camera and using one of those crappy image editing tools that come built in (on the camera) you could remove that pattern in the pixels ( as wikipedia puts it “To obtain a full-color image, various demosaicing algorithms can be used to interpolate a set of complete red, green, and blue values for each point.”) thus making it hard for the program to detect change in the pattern.
It’s not as easy as I had first thought BUT I think it would be possible. I just get real skeptical when people start talking about the capabilities of computers as if they are finally smarter then us (It will happen some day). The softwares strengths seem mostly based on it’s ability to detect patterns in the RGB layers of an image. If you remove ( through the demosaicing mentioned above) the pattern you remove the program’s abilities.
I still think my best point is that if the program was infallible they wouldn’t have to mention that they look for rookie errors like misplaced shadows. Once again I hope no one cheats and everything is legit and I am bringing this up for the sake of discussion not to inform people who would try and cheat.
And for those who want to discuss how my name is spelled I am moving that thread onto the discussion board at anonymousinsultsfrompeoplenotinterestedindiscussion.com.
Nick said this on June 20th, 2007 at 11:02 am
Scott,
First of all I don’t think you are an idiot and it’s unfair to peg you at that by asking questions (though I do admit offering to cheat for hire lacks some tact). After all that is how we learn. I am an EEng who works with DSPs for medical imaging so please forgive me if I start going into jargon and such, being surrounded by engineers takes a toll on language but here goes….
Here are a few things you should keep in mind.
Like any analog conversion process is a certain amount of random noise that comes off of a CCD, the noise is pretty distinct in that it’s very hard for software to recreate it. Editing an image will create artifacts within this noise. Adding the image back to the camera will not put the image back through the A/D conversion process again.
JPEG compression uses a quadratic table to control the quality and compression of the image, these tables are often very unique and can be tracked to the camera or software package. How Photoshop compresses a JPEG is very different (relatively speaking) to how your Sony Cyber-shot does.
What we see of an image is just one way of looking at the data in an image. There are many ways of looking at the data and many transforms that can be applied to the data that make things a lot more evident. A simple example of this (though not all that useful for these purposes) is the histogram in Photoshop. This can get very complex and involve some of the worst math has to offer but in exchange for the complexity can do some pretty amazing things.
I am not expert in digital forensics but I am pretty familiar with the concept, so hopefully there might be some insight within this. Question or comments please email me at nicklinn at eagle dot ca
Link Sharing: 6/20/07 » Webomatica - tech, movies, music blog said this on June 20th, 2007 at 11:20 am
[…] Fishing Site Uses Technology To Catch Those That Stretch The Truth: Literally. […]
sammy said this on June 20th, 2007 at 9:35 pm
how about you JUST CATCH A BIGGER FISH?
Lewis said this on June 21st, 2007 at 12:25 am
I’m with Scott, i’m sure it’s possible to fool the program, not easily but always possible. To recreate the RGB camera thingy why not print the edited photo of your fish and take a photo of that (not the best method by far, may look crap, but it’d work). Also what’s to say you couldn’t design your own (rotation/resizing) algorithm completely different from current ones that the program couldn’t detect or apply this RGB camera light thingy with a filter afterwards?
lawton said this on June 21st, 2007 at 4:07 am
wow! that is interesting that law enforcement is using this software. Great site and thanks for the PS tips. I will tell my readers.
-Lawton
Bounty Fishing: 5 Ways to Easily Spot a Photoshopped Image — How To Start Your Design Business said this on June 21st, 2007 at 4:09 am
[…] read more | digg story Sharing Is Good. These icons link to social bookmarking sites where readers can share and discover new web pages. […]
Hardcore said this on June 21st, 2007 at 4:40 am
Re: Jonah said this on June 20th, 2007 at 8:57 am
Here is how it works
1- After your images are uploaded to our database, they are imported into the BountyFishing Forensics software.
2- The software first detects the image’s authenticity to ensure it is the original photo.
3- Using the human eye and other reference points in the photo of you holding your catch, the software verifies the accuracy of your ruler.
4- Finally the length of your fish is measured and validated from the image of the fish and ruler.
Compete said this on June 21st, 2007 at 4:46 am
Answer to what John Lehmann said on June 19th, 2007 at 1:23 pm is found below. Wake up!
http://www.bountyfishing.com/enterprise/howto/compete/
Lake Ontario United Partners with World's Largest Online Fishing Derby - BountyFishing.com » Lake Ontario Fishing Forum Blog Articles said this on June 21st, 2007 at 6:51 am
[…] One of the cooler uses of forensic software technology is used to ensure that any image submitted to the site has in now way been altered via the use of Photoshop or any other image editing program. This was of course many people’s biggest concern and it is nice to see that BountyFishing has worked closely with Hany Farid, a professor at Dartmouth College who is renowned in digital forensics. Greater detail about digital forensics can be found here. […]
Tofu said this on June 21st, 2007 at 11:25 am
Re: Scott Schomburg
I haven’t read through all the posts in this message board. Suffice it to say I don’t think you actually read Prof. Farid’s papers on how the detections work.
After skimming some of the white papers I grasp what the mechanics are and I think the theories are sound. I just don’t like the samples the researchers used. The cloning samples are very obvious in some instances. I would be very interested to have you submit a non-manipulated and manipulated version of your work to see if the software can “catch” you.
BTW, if you really are a Photoshop aficionado I encourage you to check out PhotoshopContest.com some time. We’d love to see your work.
Cheers!
Motorcycle Guy said this on June 21st, 2007 at 2:54 pm
Wow who would’ve thought this technology would be applied in this way?
peter said this on June 22nd, 2007 at 6:48 am
Scott said I am now offering my services (for a small fee) for anyone who wants there fish enlarged in Photoshop. It comes with my personal gaurantee that they won’t be able to detect an alteration. if you would like to win some fishing awards.
How about some jail time in exchange for your services
Piccix - Infinite Imagination » Blog Archive » Photoshop Forensics - How To Spot A Fake said this on July 9th, 2007 at 4:48 am
[…] When it comes to spotting fraud in digital photo’s it is not always so easy to tell the real from the unreal. When reputation and honour are at stake - spotting a fraudulent fish, say in a fishing tornament - it is time to call Photoshop Forensics. […]
kerry Francis said this on September 1st, 2007 at 8:30 pm
I have to give your site an A+ for thorougness ,honeesty and integrity
In all honesty this is the first time I have come acr both an explamation and indeed of example .
If anything in the world of the internet it usually works the other way.
Thanks for setting such a practice and en example to follow
And in all places a fishing website so to speak
hiutopor said this on September 17th, 2007 at 9:42 am
Hi all!
Very interesting information! Thanks!
G’night
Bounty Fishing: 5 Ways to Easily Spot a Photoshopped Image « Design News said this on October 4th, 2007 at 5:40 am
[…] read more | digg story […]
World Design O! » Blog Archive » Bounty Fishing: 5 Ways to Easily Spot a Photoshopped Image said this on October 24th, 2007 at 10:43 am
[…] read more | digg story […]
Mister Ian’s Weblog from Kuwait » Don't use Photoshop to cheat on fish stories said this on November 25th, 2007 at 2:00 pm
[…] Bounty Fishing Blog » Photoshop and fishing don’t go hand in hand […]