How to find stolen photos on the web

Discussion in 'Business of Photography' started by cm|1, May 12, 2010.

  1. Most photographers just learn by coincidence that their images are used illegally. It’s a shame because professional photographers and semi-pros need all the money they can get these days.
    There is a new service that automatically scans the internet and reports all findings weekly:
    Detective Happenstance can retire now… bad luck for people that use our photos without permission :)
    In contrast to Tineye it works automatically and sends a report once a week. With Tineye you have to sit in front of the computer and work. I tried it. Results are pretty nice, they found several of my images in the first week. I think they concentrate a lot on media websites, corporate websites and blogs and scan more and more.
  2. The pricing ranges from US$10 a month for a minimum service to $40 a month.
    I'll confess I'm always happier when a post pushing some commercial service or product is accompanied by a disclaimer of personal interest on the part of the poster.
  3. Sorry if you get the wrong impression - I am not paid for this. My personal interest was to find my photos, and it paid out, that's all.
  4. Not at all, Clemens, it's just that sometimes posts like yours turn out to be a plug from the author of the program or some such, but not so identified.
  5. I personally use Tineye. It's free and works pretty darn well!
  6. it


    Tineye works well, but only indexes a tiny % of what's out there.
    Personally I think the whole internet photo theft scare thing is way overblown. Most people who are overly worried about it don't have images worthy of theft.
  7. Spearhead

    Spearhead Moderator Staff Member

    Personally I think the whole internet photo theft scare thing is way overblown.​
    100% agreement. The only things I don't put up now are ones with news value, because those do have time-based value.
  8. Painful theft is when they take away something from you, and you no longer have it.
    Copy of an Internet image is rather your recognition as a good photographer. If you post an image in public view you are inviting theft. Theft would be if they make profit at your expense.
    Offering a service to find your stolen photos ? - is a good idea, that could also be exploited by the thieves. Once they start paying to find, it will make sense to steal even more.
  9. I found at least 6 blogs, mostly chinese and turkish, that stole my pics here from That are 34 pictures that are given from blog to blog and I have no chance to do anything against it. They earn money with ads and attract people to visit their site with my pics. That is the unfair truth. I have removed all my pics from and all similar sites. That was my first reaction which I do not like in the long run. But I still do not have any real solution for that. I wish I could infect my pics with a virus that works only when pics are stolen:)) (Ironic mode switched off)
    Any ideas?
  10. I have a question ..... How do these services find the images? Do they have to be taken from a website? Are they comparing image file names? If an image is scanned from a print and then used on a website, will it show up as stolen?
  11. Almost certainly, each company has their own proprietary method to find similar images. Some claim to, and do find images whether or not the image has been resized, color corrected, rotated, etc. At the current state of development, the degree of alteration that can be tolerated is probably not large (...see last paragraph, below).
    "Similar image" services crawl the web, and reduce each image they find to a set of numbers called the feature vector, which they hope characterize the most important aspects of the image. There are many, many ways to do this, but, in general, if this is well done, it can be a very time consuming process. Simplified algorithms can be used to speed up this part of the process, but the results won't be as accurate, ie, more false positives (ie, images they claim to be similar), and more false negatives (ie, images that really are similar, but which don't show up in the results of their searches).
    When a user submits a base image with the hope of finding similar images, the search service computes the same set of numbers (ie, a feature vector) for this base image. It is then a very fast and relatively simple arithmetic procedure to compare this feature vector to the all feature vectors stored for the images they have extracted and processed when web-crawling.
    I doubt that many of these "similar image" search organizations base their hits on just file name or keyword similarity, although my guess is that once a set of visually similar images have been found, if any of these images have similar file names or keywords to the base image, they will almost certainly promote such images to the top of their list of similar images.
    It's reasonably easy to do an independent test of the efficiency and accuracy of these "similar image" search engines. Just post a set of modified images on a popular website that these engines regularly scan. Wait a few weeks until the images are scanned, and then submit the unmodified image in a search query. One can use this method to test exactly which modifications of the image make it sufficiently dissimilar to the original that it is not caught.
    I didn't have the time or inclination to do a serious test of this sort, but a couple of months ago, curiosity got the better of me and I did such a test, but only with one single image. I did a L-R flip, added some perspective (keystone) distortion, changed the background color a bit (less than about 25% of this image), added some light vignetting, and down-sized it modestly. I did absolutely nothing to the colors or brightness of the main subject. Any person shown the before and after versions of this image would *immediately* spot the similarity. However, not one of the three or four "similar image" search engines that I tried caught the modified version (but most of them did find a match to the original, unmodified version).
    Tom M

Share This Page