17 April 2012

Cyberspace May 2012


Lies, damned lies


In 2050 no grandchild will see a photo of his grandfather sitting on his antique motorcycle, because that photo was taken on a phone or digital camera and was never backed up or handed over to the children (“Here honey, take my 30 GB of family photos before I die”).. Another problem with digital photographs is that they are easily edited... but are those edits undetectable?

You might be involved in an AVO defence, a claim against police or a family law matter where some photographs are being tendered. What can you do to ensure that they haven’t been tampered with? I’d start by cross-examining on the chain of custody of the digital images, starting with the photographer and ending with the person tendering them in court.  The concept of an “original” photograph is fairly nebulous - perhaps the only original is that on the SD or Compact Flash card - everything else is suspect. But the truth is that you can analyse a photo that has been resized, cropped, altered and find the fakes.

There are many techniques (and multiple techniques should always be used) but only some deal with visible issues. Classic visible problems are where the light appears to illuminate a subject from several directions when there clearly could only have been one light source. This analysis can show that one or more subjects have been added, moved or reversed. Another visible problem is where perspective anomalies arise.  If an object is inexpertly added then its perspective will not match the rest of the photograph. This can show, for example, that the wheels on a car are too close together or people are too far away from a background object. Changes in highlights (bright areas) where you would expect them to be similar was noted in a Scientific American article (http://goo.gl/Og0Tv), where a photograph of American Idol judges was analysed to show it had been doctored.

But what about edits that are seriously professionally done using quality software? These are still detectable.  A great recent example was by Dr Neal Krawetz, who has been conducting digital photo forensics for many years. In the USA a recent lottery draw for over $640M was world news. A person posted three photographs of a “winning ticket” on Reddit, and Dr Krawetz decided to examine them (http://goo.gl/qhcjS). . These photographs were seriously believable visually, but the context indicated they were probably fake, and he ultimately proved so. How?

The first picture was analysed to see if different areas of the image had been compressed at significantly different levels (all JPEG photos have some degree of compression). Even after multiple saves we should see consistent degradation across an entire photo. This sort of analysis will easily reveal that something has been added to or removed from a photograph, but if something has been copied within a photograph then other tools will be required.

The next step was to consider whether tools such as Photoshop were used - these introduce distinct artifacts that are peculiar to the brand of software used. After processing experts can visually identify which software has been used.

Another anomaly that can be introduced is varying colour spaces, which is what the Dr used to detect the lottery fake. Altered parts of the photo will be revealed by detecting changes in the colour values used in different parts ofthe photograph.

The moral is that you need not accept defeat in the face of what your client tells you is a digitally altered photograph - you should consider calling in the experts.

No comments:

Post a Comment