AMITIAE - Tuesday 22 July 2014

Cassandra: Identifying Images on Social Networking Sites Using Sensor Pattern Noise (SPN)

apple and chopsticks


By Graham K. Rogers


A few months ago, some research was published that put forward the idea of making identifications of suspects or the locations they had visited, by the use of enhancements to reflections in the eyes of subjects in photographs. I examined this at the end of December last year and in the item I wrote then commented on this using a number of my own images in which reflections might be used for such purposes.

metadata While some metadata could be used for identification of camera or user, this is easily removed in post-processing, so may not always be available.

Some recent research, which was presented at a conference on Computer Vision Theory and Applications (VISAPP) in Lisbon recently, has looked at the way modern digital cameras may have unique identifying characteristics, particularly noise in the camera's sensor element. This is called Sensor Pattern Noise (SPN).

The authors Riccardo Satta and Pasquale Stirparo explain that light goes through the lens, then through an anti-aliasing filter. It reaches reaches after passing through a Colour Filter Array (CFA). There are more steps after the senson including de-mosaicking and post-processing. They explain that "Each step of this pipeline may leave artefacts on the image that can be used as a signature of the camera device."

The paper details the considerable amount of research already carried out in this field since 2006 and points out shortcomings in some of the consideration of ways that artefacts might be used for identification. For example, some depend on algorithms used by manufacturers. These are usually model specific and not unique to a specific camera.

Dust mainly affects cameras with interchangeable lenses and could identify a single device. This may not always be valid, however, especially as lenses may be switched between cameras and even sold on; and dust deposits change over time (most photographers will also clean the lenses). However, specific lens "aberrations" may help in identifications. This may be more useful in terms of smartphone use as lenses are not interchangeable.

However, research has shown that Sensor Pattern Noise has particular "characteristics of uniqueness and stability". Once a particular set of characteristics have been identified using the algorithms developed, they can be matched against images, perhaps including those uploaded to social networking sites.

As was found in the case of Mr Swirly Man (Christopher Paul Neil) who was finally arrested in Bangkok, once identifications are made, finding the persons and questioning them are likely to reveal more information and bring a case to a conclusion.

Experimental results showed that despite a number of variables - including compression, ISO settings and image alteration - the use of the SPN is a feasible method to use in identification and tracking. The researchers used 10 Flickr and Facebook accounts in their research, each with several images (up to 192 max).

While identifications did not approach 100%, which would be desirable for forensic accuracy, the probability (above 50%) would point to the need for further inquiries, including matching information from other sources.

The researchers were particularly interested in the use of this means of identifying offenders in cases of "On-line Child Abuse, cyber- bullying, or thefts of smart-phones." Further research will seek to enhance the ways in which identification by means of SPN could be used.

See also:

On the usage of Sensor Pattern Noise for Picture-to-Identity linking through social network accounts (Riccardo Satta and Pasquale Stirparo).

Graham K. Rogers teaches at the Faculty of Engineering, Mahidol University in Thailand where he is also Assistant Dean. He wrote in the Bangkok Post, Database supplement on IT subjects. For the last seven years of Database he wrote a column on Apple and Macs. He is now continuing that in the Bangkok Post supplement, Life.



Made on Mac

For further information, e-mail to

information Tag information Tag

Back to eXtensions
Back to Home Page

All content copyright © G. K. Rogers 2014