How does TinEye actually work to identify a photo (does it useEXIF?)

Discussion in 'Digital Photography' started by William Don**ly, Dec 27, 2012.

  1. How does TinEye actually work to identify a photo (does it use EXIF?)
    http://www.tineye.com

    A friend told me I can search TinEye for free to find my pictures
    to see if anyone has posted them to the net.

    I tested a few of mine - but none showed up on the net.

    But maybe people removed the EXIF information or cropped them or filtered
    them in software or did something to obfuscate them.

    I doubt you'd have the exact algorithm (nor would I know what to do with it),
    but, do you at least know roughly how the TinEye program discerns duplicates?
     
    William Don**ly, Dec 27, 2012
    #1
    1. Advertisements

  2. TinEye doesn't need EXIF data. I'd imagine that they analyze images for
    simple features that can be listed and cataloged. The features for an
    image doesn't need to perfectly describe it. It just needs to narrow
    down potential matches enough that they can be further analyzed in
    real-time. It's not unlike the secret-sauce complex analysis that web
    page search engines perform. They're not matching words, but meanings
    in a context for a specific culture. Google has claimed that they can
    even interpret the images in a web page.
     
    Kevin McMurtrie, Dec 27, 2012
    #2
    1. Advertisements

  3. William Don**ly

    DaveS Guest

    Here is my example:
    http://www.tineye.com/search/4e2829463c185c2b8bdea75584853eb3c9ba5077/

    The first image is my own. Three others seem to be taken from the same
    viewpoint.

    I thought at first this tool could identify copied images, but now I
    think it just identifies similar images.

    Dave S.
     
    DaveS, Dec 27, 2012
    #3
  4. William Don**ly

    sobriquet Guest

     
    sobriquet, Dec 27, 2012
    #4
  5. I have no more idea than David. "image search algorithm" brings up some
    pointers on the first page. A deeper dig finds one paper comparing three
    approaches. I'm sure you'll be able to turn up more with research.

    A Comparison of SIFT, PCA-SIFT and SURF by Luo Juan & Oubong Gwun
    http://www.cscjournals.org/csc/manuscript/Journals/IJIP/volume3/Issue4/IJIP-51.pdf
     
    Charles E. Hardwidge, Dec 27, 2012
    #5
  6. Pattern recognition software.
    Its the same way the government tracks your movements via
    facial patterns from 100 meters away.
    Tineye could not possibly download each file then compare
    the exif data from each one.
     
    Paul in Houston TX, Dec 27, 2012
    #6
  7. So that makes at least 3 of those picture finders out there:

    1. Images.Google.Com (finds similar pictures also)
    2. TinEye.Com (finds the same picture)
    3. yandex.ru (not sure what it does since it's in Rusky)
     
    William Don**ly, Dec 28, 2012
    #7
  8. William Don**ly

    sobriquet Guest

    It's a bit doubtful how effective these kinds of attempts to
    police your images are likely to be.

    Take for instance these beautiful pics of slime molds by a flickr
    user:

    http://www.flickr.com/photos/myriorama/58762793/in/set-1271006/

    This user seems to object to downloading of his images (downloading
    is disabled at flickr). Yet if you search for them, they pop up on
    many sites:

    http://tinyurl.com/d3m6nlb

    Some of them even seem to neglect mentioning the source:

    http://metapsychic.tumblr.com/post/4017720811/slime-mold


    It seems like a rather futile effort to try and prevent pictures
    from being reproduced and distributed.

    But perhaps there are companies that can be hired to police your
    intellectual property.
     
    sobriquet, Dec 28, 2012
    #8
  9. William Don**ly

    J. Clarke Guest

    I listed a New York license tag on ebay the other day. I fed that
    picture into Tineye, it did not find the one I posted, but it did find a
    bunch of other New York license tags with different numbers. I also fed
    it the original of a photo that I have had up on flicker for more than 6
    years--it did not find any matches for that one. It's clearly working
    by pattern recognition. It also clearly does not have enough of the
    images on the Web indexed to be considered a reliable tool. This is not
    surprising--Tineye claims to have about 2.3 billion images indexed.
    Flickr alone claims to have more than 6 billion online.
     
    J. Clarke, Dec 28, 2012
    #9
    1. Advertisements

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.