Why we need an imagery analysis expert more than ever – and not just because of AI

“Many people believe they can accurately interpret visual evidence, when in reality imagery can be highly misleading without proper technical understanding.”

Almost everyone now carries a mobile phone equipped with a camera and is willing to record whatever they observe. When this is combined with the rapid growth of CCTV systems, dashcams, doorbell cameras, and recorded conference calls, video and still imagery now play an increasingly significant role across a wide range of investigations, dispute cases and personal injury claims, in both civil as well as criminal contexts.

However, widespread exposure to television, film and online media has created a false sense of expertise among viewers. Many people believe they can accurately interpret visual evidence, when in reality, imagery can be highly misleading without the proper technical understanding. As a result, there is a real need for expert analysis to accurately determine what imagery truly shows, rather than what it appears to show at first glance or, more importantly, what the viewers confirmation bias wants it to show.  This has been clearly demonstrated by the diverse interpretations circulating on social media in response to videos covering major news events.

Common areas of misunderstanding include camera lens effects such as foreshortening and perspective; the limitations of image resolution; motion blur; frame rate; image compression artefacts; the characteristics of monochrome imagery; and the influence of cognitive factors such as confirmation bias. Without specialist knowledge, these factors can easily lead the uninitiated to draw incorrect or unsupported conclusions from visual material.

Desktop image processing and editing software is now freely available and easily accessible to the general public. As a result, it is increasingly common for untrained, nonspecialists to attempt to enhance or analyse imagery without a full understanding of the underlying processes or potential limitations involved.

Improper use of enhancement tools and filters can introduce unintended artefacts, distort fine detail, and alter the apparent content of an image. Rather than clarifying evidential material, such untrained manipulation can obscure or fundamentally distort the original data, potentially leading to inaccurate interpretation and unsound opinions and conclusions. Going as far back as 1988 in the case of Edward Browning, convicted of the murder of Marie Wilks on the M50 motorway, his High Court appeal determined that incurred processing had distorted CCTV imagery of vehicles passing a traffic camera making accurate vehicle identification impossible.

While modern digital video recording systems represent a substantial advance in terms of image quality, storage capacity, longevity and accessibility, they also introduce new and less obvious challenges for interpretation. Contemporary systems employ a wide range of recording formats, frame rates, and picture resolutions, some of these formats are proprietary to the system manufacturer and others are industry standard. Difficulties commonly arise when imagery is downloaded and transcoded into formats that are more easily viewed or shared. During this process, it is not unheard of for video frames to be dropped, playback speed altered, or the aspect ratio distorted.  Such changes have the potential to undermine the accurate interpretation of visual detail and recorded events. 

Quite apart from this, and increasingly relevant, are the challenges now posed by ‘deep fake’ content generated by artificial intelligence. Well publicised incidents, such as senior executives being deceived into authorising substantial financial transfers following highly convincing deep‑fake communications, illustrate the growing sophistication and real‑world impact of such technology. There is little doubt that distinguishing authentic imagery from manipulated or synthetic content is becoming increasingly complex especially where the creator knows what the authenticator will be looking for and how to mask these signs. Generative Adversarial Networks (GANs) use a generator and a discriminator, in other words a forger AI and detective AI, to compete against each other in a zero-sum game to produce highly realistic synthetic data. 

Given these considerations, avoid taking images at face value, if the interpretation matters, seek input from a qualified and experience imagery expert.