Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

Implement a web-based app which performs copy-move/clone-stamp detection in images. Similar to the experiment here: http://coding-experiments.blogspot.com/2009/03/detecting-cop...

So, the web-based version (kept simple) would allow a visitor to: 1. upload an image (& enter their email) 2. process the image in the background (the detection algorithms that I've seen aren't fast) 3. email the visitor when the detection is complete, with a link to results.

This would be a valuable service to newspapers, magazines, and news websites–who must make sure that news/photojournalism images aren't tampered with in Photoshop.

If you could improve upon the algorithm linked above and open-source it along with the web app, I think it would be a great resource. Of course, you'd still have plenty of room to build a for-pay service at a higher level. I know larger news organizations that would pay to be able to run several images through such a service and save the results.

Feel free to get in touch if you'd like more details.



That is a pretty interesting idea. Are there other photo tampering detection methods other than the obvious clone stamp? For instance, does color correction leave a fingerprint that could be detected via some sort of probabilistic or algorithmic way?


I don't know the answer, but basic color correction isn't really a cardinal sin like pasting/clone-stamping is. Performing white balance or converting to B/W is usually fine (after all, a lot of newsprint images end up B/W).

Here's some examples where this behavior has been uncovered: http://www.poynter.org/content/content_view.asp?id=28082

http://www.nytimes.com/slideshow/2009/07/05/magazine/2009070...


Color correction of an 8-bit-per-channel image will leave gaps in the color histogram. It's usually possible to conceal those gaps, though, by blurring the image slightly or adding some noise.


tineye.com doesn't do all of this, but it will sometimes match different photos of the same thing / location.


Yeah, Tineye looks cool. Slightly different application however.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: