Detect edited photos with izitru

[colonel_mortis]

Source: http://www.theverge.com/2014/5/7/5692010/izitru-image-analyzer-detects-modified-photos

More information and official site: http://www.izitru.com/

 

Izitru, pronounced a bit like "is it true," lets you upload a photo and automatically get a computer's analysis of whether or not it may have been modified. The results are simplistic, generally telling you that it believes an image to be an original — as in, it's come straight from a camera — or that an image may have been modified, meaning it's been saved again since coming out of the camera. That means photos flagged as potentially being modified span a wide range of possibilities. It isn't only complex edits that Izitru might flag, but also photos that have been saved for a second time, even without any actual edits to them.

Essentially, izitru is an image sharing site (like imgur) that validates whether the image is genuine or not. Although the Verge article implies that it is only based off data which identifies whether it is the original image from the camera, the video on their site (linked above) suggests that it also performs further validation to see whether it looks genuine or not as well.

Although it's obviously not going to be 100% accurate, and it will definitely get loads of false positives (rotating an image resaves it, and would therefore mark it as suspicious), it's interesting to see the advances in forgery-detection. Although it's still behind the techniques used to fake the images, I look forward to seeing how accurate a service like this will get in the future. Perhaps at some point we will actually be able to trust a photo again (unlikely but possible).

[Snickerzz]

~the following is based off what i learned from a tv show~
couldnt you look at the sub pixels and see if they all came from the same place?

[colonel_mortis]

~the following is based off what i learned from a tv show~

couldnt you look at the sub pixels and see if they all came from the same place?

To an extent, you can analyse the pixels to see if there are any anomalies which look like they have been pasted in, but as far as I know, it's fairly tricakble as long as you know what you're doing (and have time to spare faking it). I guess this site will do some form of similar analysis of the image, although purely automated so not particularly accurate. There is no data in the subpixels apart from their colour to help validation (although I presume that's not what you mean).