Visual Search Engine Tracks Stolen Images

Barence writes "A new visual search engine could help photographers keep track of their photographs whenever, and wherever, they appear on the internet. The TinEye search engine allows users to search by uploading a picture rather than typing in a keyword. It then conducts a pixel-by-pixel search across the internet, flagging all instances of that image even if it's been cropped, merged or digitally altered in some way. It's not just for copyright enforcement though; 'it's being used by researchers who need to find where an image came from to provide attribution, even people who are trying to find out who people are in old photos.' It's currently in beta, but you can try it out."

Kind of Misleading on the Old Photo Identification

"it's being used by researchers who need to find where an image came from to provide attribution, even people who are trying to find out who people are in old photos."

This may be nitpicking but I read the FAQ and it does not, in fact, claim to be able to accomplish this unless that exact same 'old photo' is posted elsewhere on the internet:

Can TinEye find alterations of a query image?

Yes. As long as they are alterations of the same query image, TinEye can find them and include them in your search results.

Note that search results are ordered by 'relevance' (i.e. how well the result images match your query image), so image alterations are typically found at the end of your search results.

How does TinEye work?

TinEye uses sophisticated pattern recognition algorithms to find your image on the web without the use of metadata or watermarks.

TinEye instantly analyzes your query image to create a compact digital signature or 'fingerprint' for it. TinEye searches for your image on the web by comparing its fingerprint to the fingerprint of every single other image in the TinEye search index.

So this example they list of the soldier must rely on the fact that the website contained the same exact image that the people had of the old soldier they were looking for. I can't expect it to take any image of Person A and return every single image (past & present) of that person. That's ridiculous.

I would expect that to work out very infrequently as I'm not aware of any huge digitized databases of old photos or even newspaper microfiche. Hell, I have postage stamp-sized photos of my grandparents with people who nobody knows who they are. I don't think this tool could help me.

Re:Kind of Misleading on the Old Photo Identificat

"it's being used by researchers who need to find where an image came from to provide attribution, even people who are trying to find out who people are in old photos."

I think in this context, it's pretty obvious that the software's not trying to discover who people are, or who shot the photograph. It's the researchers who use this tool. If you have one website without attribution or other names, and you search for other pages, you might find a different page that has the same image along with more information.

Re:

Now what would be handy would be if it could somehow sort them chronologically (maybe using the metadata, or maybe if the server will give the date-modified on the picture...). That would reduce the amount of searching if you knew you were going for the oldest known copy, e.g. you wanted to know where it originally came from + whatever info there was about the picture that might not be quoted elsewhere.

Re:

You have to login to read the FAQ. If anybody wants to avoid jumping through the hoops, here's the FAQ as a gif [imageshack.us]. Sorry about the resolution, you'll just have to pick a good zoom level...

I think everybody understands...

that the real purpose for this is to find the rest of sets ;)

Re:

It might be better if it were able to find approximate matches, or face match & find. Still, you've gotta start somewhere ;)

Embedded Codes

Images can also have an embedded code, i.e. steganography [wikipedia.org], which could possibly be used to speed up searching. This would allow the web crawler to know exactly which part of the image to look at to determine if it matches the key the crawler is looking for, rather than a brute force pixel by pixel search.

Unfortunately...

That sounds like a great idea... from here --> forward (for those who start the steganography). Of course, being able to search without embedding extra stuff is more useful... still, more speed is better.

Re:Embedded Codes

"rather than a brute force pixel by pixel search"

They're blatently not pixel by pixel comparisons... look at the tech, don't listen to the woman! If it was pixel based then an image saved using two different implementations of jpeg wouldn't match up. It's probably more likely that a map of lines, shapes, patterns etc in the image is built up, and then they are what's compared. This means images that are different sizes, have different light/colouring (such as a high quality scan vs poor quality) and colour depths, but are of the same thing, can still yield results.

Err... or is that not what you meant by pixel by pixel search?

Re:

Well, they don't actually do a pixel-by-pixel search... they index them and create digital fingerprints, then it does a pixel-by-pixel on your "search query" image and compares its fingerprint with the fingerprints in the database. Pretty neat.

Funny thing, but I just shifted a bit a pixel.

The least significant bit of each pixel. Oh, and now it appears that this tool doesn't work. (At least, I would suggest it isn't that good, I could be wrong. The article appears to suggest that it is that good, if you can take a photo on your phone of a painting, and then find an article on that painting...)

Oh well, I guess people still haven't learnt that the old ways of copyright are only hanging on through inertia.

Oh, and queue the predictable (and correct) responses about how you can't "steal" digital images. To steal a photo or a picture, you would have to take a physical copy belonging to someone, and deprive someone else of that physical copy, without their permission. (And the word "steal" doesn't appear to appear in the article, added to provoke page views I guess.)

Re:Funny thing, but I just shifted a bit a pixel.

The least significant bit of each pixel. Oh, and now it appears that this tool doesn't work.

Yeah, how about you just watch the video on their website before suggesting that what they do would be as retarded as comparing the values of each pixel. It's surely closer to cross correlation, meaning it's nothing like comparing pixel values but more like correlating the image's space-frequency components.

By the way, does anyone have any clue what information they store and compare? They obviously don't cross correlate your search image with every image in their index every time you search, so what could they possibly store that would allow them to correlate images?

Re:Funny thing, but I just shifted a bit a pixel.

Oh, and queue the predictable (and incorrect) responses about how you can't "steal" digital images. To steal a photo or a picture, you would have to take a physical copy belonging to someone, and deprive someone else of that physical copy, without their permission according to SlashDot, but not the English dictionary.

Pet Peeve of mine: That's not the definition of "steal". It's only the SlashDot conventional wisdom. It's really not that hard to look up words on the internet. Here's a link to a dictionary [m-w.com].

Steal:
1 a: to take or appropriate without right or leave and with intent to keep or make use of wrongfully

Appropriate:
  3 : to take or make use of without authority or right

This impressed me...

I found quite a different result. I nabbed an old photoshopped pic I did a few years ago, and uploaded it. TinEye came back with two results, being the two source images from the photos. That's impressed the hell out of me.

Gatesfeld search results [danamania.com]

For the full size photoshopped version, Gatesfeld [danamania.com] if you want to try the search yourselves.

Re:

Yes, there is a difference, but it's a moot point - both works required your time and labor to create, the book obviously takes a LOT longer to create, yet a book, a paperback, sells for a tiny fraction of the cost of a well crafted, hand-made chair.

You focus on the one completely irrelevant part of the discussion: the time spent creating the book vs. the chair. It's completely irrelevant because "book" is not essential to this discussion. You can replace "book" with any other copyrighted creation, including one that took about 1 minute to "create".

Copyright only makes fabulously wealthy (by and large) those artists who have created compelling material... enough so that many people are willing to purchase, at a fraction of the cost it took to make the original product, the work of the artist. They're buying labor and talent and skill, which is the same thing you're buying when you buy a hand-made chair (I'd hope, anyway, as hand-made furniture tends to be quite expensive).

Not really, not any more. These days advertising and exposure has a lot more to do with it than how "compelling" the material is.

No, you're paying for the labor, skill, and talent... and people who put more effort, skill, and talent into their creations are rightfully awarded with more than those who put little effort into their work.

No, you're paying for the advertising. For an example, there's lots of bands o

What would be really cool

What would be really cool is if you could upload a transparent 1x1 pixel image and it returns every image on the internet

Yes

Re:

What would be even cooler is if you could search for transparent 1x1 pixels.

Logo hunting

So to test it out I grabbed a couple of logos (AIG, Slashdot, Bluesquare, Nike swoosh) and found that what it will do is find scaled down images or ones of lower quality but it won't handle significant colour shifts. So AIG for instance have a blue logo but sponsor Manchester United where their logo is displayed on a red background, the Nike swoosh I tested had a white background and all I got was basic black on white swoosh elements.

Now with photos this is less of an issue as major colour shifts are unusual but it does mean that for commercial and design art its not really as applicable.

What a great idea

I should create a page for movies and mp3s too, a place were directors and producers can upload their content to see if anyone has copied it already!

Terms of Service:

By using Idée's TinEye website you signify your agreement to the following terms and conditions, which may be updated by us from time to time without notice to you.

Submission of pornographic or illegal files is strictly prohibited. Do not submit any file that can be construed as pornography or is in violation of any law.

No porn searches?

::Blows a raspberry::

Failure to comply with these terms may result in termination of your TinEye account at any time, without prior notice and at Idée's sole discretion.

Ahhh.... okay.... don't search for porn, or we might not let you search for more porn from that particular account. Gotchya. Hehe.

-

What is pornography?

Submission of pornographic or illegal files is strictly prohibited. Do not submit any file that can be construed as pornography or is in violation of any law.

How the hell am I supposed to know what their company considers pornography? Can I search for The Joy of Life by Henri Matisse? [artquotes.net]

The company is based in Toronto rather than some ultra-conservative U.S. state; that gives me an epsilon more confidence the company won't take the "nudity = pornography" stance. Still, I wouldn't be surprised if a search equivalent to a risqué ad campaign in Europe would get you banned.

Re:

That does raise the question though... how are they determining who is searching for it? Is it based on the results? Image Z returned mostly results from sites whose primary keywords were "porn" and "XXX" so this image is probably porn. Do they have a human just randomly check? or check the ones that the aforementioned programmatic filter found? It seems like there'd be a lot of room for error...

robots.txt

My robots.txt excludes access to my huge collection of images.

So, either one can prevent discovery by this tool in a very simple way, or it ignores robots.txt. Which is it?

Re:Small letters

I suggest that if you're not going read TFA and can't be bothered to do any further searching of the site you are slamming that you read this page:

http://tineye.com/terms [tineye.com]

which specifically states the opposite of what you claim might be there.

The letters aren't that small.

Million dollar ideas

I've had many great ideas that somone else had already implemented stage 2 before I could:

• Drive through window on doughnut/coffee shops. I saw my first one about three months after I had the idea.
• Building a universal remote control into a Star Trek phaser or tricorder housing and selling them at conventions. That idea died as soon as I googled it [redferret.net].
• Adding a button on your TV that makes your remote control beep, so you can find the thing when it decides to walk away. Again [keyringer.com].

Steps 1 and 2 of your plan are n