Passive Image Forensic Techniques for Source Identification
- When?
- Thursday 18 November 2010, 11:00 to 12:00
- Where?
- 39BB02
- Open to:
- Staff, Students
- Speaker:
- Mr Phil Bateman
Recently, much interest has developed in identifying reliable techniques that are capable of accurately ‘uncovering the truths’ regarding the pre and post- processing of a digital image, without the requirement of actively injecting a digital watermark or signature into the image data. Whilst watermarking schemes have been shown to be useful for protecting the integrity of the image, there always exists the underlying risk that the watermark data might be forcibly or accidentally removed. When this happens, the image is effectively stripped of its identity, and its integrity is extremely difficult to prove. Forensic techniques aspire to achieve similar objectives but do not rely on the strength of embedded data. Instead, the ambition is to identify the facts of an image, based solely on the data provided.
In this research, we focus on two key areas of the Image Forensics field: camera identification, and forgery detection. From a legal perspective, there exist several scenarios where it would be useful to successfully prove the origin of an image. If illegal images are found on a suspect’s computer, it would be useful to ascertain whether or not those images were also captured by the suspect. Similarly, there are many applications for accurately detecting content manipulations that alter the integrity of the original scene, since digital images are often presented in the court of law as supporting evidence.
Our work begins with the camera identification problem, with the novel use of Statistical Process Control (SPC) as a tool for identifying anomalies in the image acquisition process of digital cameras. Control charts are used to illustrate the overall level of control associated with several devices (models include Apple iPhone 3G and 3GS, Nokia N97, and Leica D-Lux4), which are in turn reviewed in accordance with the Western Electric Rules for identifying assignable causes for the observed variation. X-Moving Range, Exponentially Weighted Moving Average (EWMA), and Cusum control charts are used to highlight the variation for a subset of the devices. By implementing such a statistical model, the forensic investigator is much better positioned to understand the behaviour of a particular device, and is ultimately able to identify the most unstable feature of the cameras image acquisition process, thereby establishing a suitable fingerprint for matching images to their source.
