FAU researchers develop new technology to tackle fake digital images

In the age of digital photography, editing has never been easier – unwanted blemishes can be removed, contrast can be increased or the lighting of an entire image can be changed. The work of professional counterfeiters is almost impossible to detect with the human eye. This raises the question of how trustworthy digital photographs really are as evidence in court, for insurance claims or in the media. Researchers from Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU) have developed new technical methods which help to pinpoint digital manipulation easily and reliably.

When checking the validity of digital photographs, forensic scientists rely on two sources: traces of the forgery left behind by editing software and other discrepancies in the characteristics of the scene or camera. Such general characteristics may include the lighting effects in a photograph.

The research group led by Dr. Christian Riess from the Pattern Recognition Lab at FAU (Chair: Prof. Dr. Joachim Hornegger) have examined colour temperature as a new characteristic for digital photography forensics. Colour temperature varies based on the environment – for example, flash produces a different colour temperature to room lighting or sunlight. Computers can be used to analyse the colour temperature for specific image sections with great precision. If the results for different image regions do not match, it is plausible that the image has been manipulated.

Computer scientists at FAU also investigated the direction of light and have improved an existing method. Slight changes in the direction of light are generally difficult to distinguish – however, algorithms can be used to pinpoint the angle of incidence for different objects relatively precisely. If individuals are added to a photograph, scientists can be relatively sure that their lighting is different, if only minimally. Previously, only surfaces consisting of a single material such as a person’s jacket could be used for reference. With the new algorithm developed by FAU researchers, the direction of light can be analysed based on diverse materials such as clothing, skin or hair. This has clear benefits: the results are much more reliable and scientists are closer to developing methods which are suitable for practical use.

Suspicious traces from photo editing software

FAU researchers have also made significant steps in identifying suspicious traces left by photo editing software. A comparatively simple and common form of digital manipulation are copy-move forgeries: this involves editing individual sections of an image with a digital cloning tool. Over two years, the researchers conducted several tests on 15 existing procedures for identifying copy-move forgeries. An application was developed based on this research which can identify copy-move forgeries reliably based on uniform criteria.

A further area of the research focused on a method for identifying changes to JPEG photographs based on the number of compressions. If a photograph is saved in the JPEG format, digital traces are left behind. If a photograph is edited, these traces disappear – but they are only removed in the areas which have been edited. This means that images which have been manipulated contain sections which have only been compressed once by the photo editing software and original sections which have been compressed by the photo editing software and the camera. Using a series of examples, FAU researchers were able to train an application to identify sections of images which had a different number of compressions.

Digital photograph forensics is of great importance to criminal prosecution. Although digital photographs are generally not yet accepted as evidence in court, there are situations where photographs may be considered during trials. Defendants in the USA, for instance, argued that photographs containing child pornography were not real-life photographs and that they had been edited after they were taken. The prosecutors had to prove the the photographs were not edited digitally and were actual evidence of a crime.

Intelligence services are also likely have a significant interest in software which is suitable for real-life applications: many image forgeries are politically motivated and are published as propaganda or to cause harm to opponents. Methods for identifying digital forgeries are also of interest to insurance companies. Insurance claims are often accompanied with digital images, which are currently not automatically checked for authenticity. This could change in the future if a programme could be integrated seamlessly into claims processing workflows.

The research conducted by Dr. Christian Riess and his team is the topic of a television report, which will be broadcast in Germany on Thursday, 4 April as part of ProSieben’s ‘Galileo’ programme.

Addition information