1.01 billion images in 0.632 seconds?


D

dorayme

http://tineye.com/ claims to have searched 1.01 billion images in 0.632
seconds? to find three or four pics out there that are identical to one
on my hard disk (yes, I pinched a small Roger Rabbit one ages ago ...).
Should this be believed?

Awfully nice idea this site, it was mentioned at uk.net.web.authoring in
recent thread.
 
Ad

Advertisements

B

BootNic

On Sat, 14 Feb 2009 05:17:10 -0500
To find one that's *identical*, down to every single pixel? Yeah,
that's believeable. They could calculate a hash value for each
image as its being added to their database. When you upload your
own image, they'd calculate its hash value. Getting a list of
images with the same hash value as yours would then be a simple
query on an indexed integer column - very, very fast for modern
databases.

I can't imagine them doing any more complicated matching in
that kind of time though - no face matching, color space
conversion, fuzzy matching, none of that sort of thing.

I have no ideal how it's done, how many images it checked, but it is
very quick.

http://tineye.com/search/efec127c48945335a83a88a3f6acf127e2ad7896

The above example appeared to take longer to upload the image then it
did to return the results.

As I think you will see, the matches are not matches, but they are
similar in part.

I find it interesting. Very nice find.

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.9 (GNU/Linux)

iEYEARECAAYFAkmW9uYACgkQylMUzZO6jeJsRQCeNTX+BWHt0zrb7LjFj8jCYJQM
ufsAn1RLkvqCXtqnHbSJXFcz1kflhw+Z
=xVHt
-----END PGP SIGNATURE-----
 
D

dorayme

BootNic said:
On Sat, 14 Feb 2009 05:17:10 -0500
To find one that's *identical*, down to every single pixel? Yeah,
that's believeable. They could calculate a hash value for each
image as its being added to their database. When you upload your
own image, they'd calculate its hash value. Getting a list of
images with the same hash value as yours would then be a simple
query on an indexed integer column - very, very fast for modern
databases.

I can't imagine them doing any more complicated matching in
that kind of time though - no face matching, color space
conversion, fuzzy matching, none of that sort of thing.

I have no ideal how it's done, how many images it checked, but it is
very quick.

http://tineye.com/search/efec127c48945335a83a88a3f6acf127e2ad7896

The above example appeared to take longer to upload the image then it
did to return the results.

As I think you will see, the matches are not matches, but they are
similar in part.

Yes, I am sure this is right and what Sherm says. I suppose it would not
have found a closer match (with the girl draped) had it taken more time
- after all 0.014 seconds is not much time! In which case, one can
conclude it's fairly unique an image.

Let me try with with an image I use a lot in html tests (it resizes so
well!):

<http://dorayme.netweaver.com.au/pics/crimson.png>

Reply: Image too simple to create unique signature

Who can argue with that. Let me try a pic I took.

http://tineye.com/search/e7a856835a9201fd28712b1340512a37e2467c96

There is a database of their own. If they got all the images on the
internet right now, I should work out how big their hard disk (imagine
*one*) have to be! But I will resist. <g>
 
Ad

Advertisements


Top