I recently testified in court as a forensic image and video expert and, as is sometimes the case, the use of some filters to enhance images was questioned. As I have written before, there is some processing that should be entirely avoided, since it lacks accuracy and repeatability. For example, we should avoid techniques which add new information relying on data obtained by a training set, or techniques which have a random component.
Some years ago, there was a school of thought that said, only classical image processing techniques available for the analog photography can be applied to digital photography in the forensic context. What are the risks of applying the wrong processing? We are not interested in having a “pleasant” image, we are concerned about extracting information from it. The risks of wrong processing are:
- Removing existing information: for example, removing the grain in a dark image can remove also important details.
- Adding new information: for example, creating or amplifying image artifacts which may be misinterpreted as a real detail.
In this reasoning, we are not referring to details at the pixel level, but at the image semantic content. In general, if I resize an image, I add a lot of new pixels but if the processing is correct I am not adding any new relevant information.
It’s important to understand that most of the image processing techniques present a compromise: I enhance something at the expense of damaging something else. For example, if I lighten an image to show better a dark part, it’s very likely to lose details in the parts of the image that are already bright enough.
For this reason, it’s very difficult, in general, to say which techniques are good and which techniques are bad. Their applicability must be related to the specific case and the parameters used. Filters are just tools, and as such, they can be used in the right way, obtaining better images, or in the wrong way, damaging the image quality or presenting wrong information.
Because of this, it’s important not to blindly apply different enhancement and restoration filters, but to apply them in order to correct a specific defect. Similarly, the tuning of their parameters must be consistent with the amount of defect I want to correct. Abusing the filters can create images which are much worse than the original.
It is therefore important, as I’ve said many times, to work with experts who have specific experience in the forensic image and video analysis field. Who know what to do, and how to identify what has been done incorrectly.
A lot of pressure may be put on the processing done by the experts, but most people ignore that there are many other processing and possible issues happening during the image acquisition and visualization phases.
A lot of processing happens in the camera itself, from CCTV to smartphones. Unless raw image pictures are used, and this is very rare, the value of the pixels in an image are hugely dependent on the processing and encoding which automatically happens inside the device to obtain the ratio between image quality and technical limitations that the producer wished to obtain.
And then, even to simply visualize the image, there’s a lot going on under the hood. Different software can decode the image in a slightly different way which can enormously impact the final result, and a lot of image processing happens on the graphics card of the PC, on the screen, or on a projector. Just play with the brightness of the projector to realize how much the visible information in an image can be impacted by such simple tuning.
There is then the most critical part of the processing: our eyes and our brain. Different people see and want to see different things in the same image. Analyzing things in an objective and unbiased way is often very difficult unless you can measure things numerically. And in fact, avoiding and limiting the various types of biases are one of the most important aspects of forensic science currently studied.
This article, written by Martino Jerian, was originally published in Lawyer Monthly magazine. Click here for the published article.
In this version, we have released a new operational mode in Amped FIVE through a panel called the “Assistant”. The concept is extremely simple but opens a world of possibilities. I had this idea during a meeting in the US about one year ago. I hacked a quick and dirty prototype in a couple of hours and showed it to a few users at the LEVA conference to gather feedback. From then on, we worked to improve it and prepare some scripts.
Sotiris, tell us what you are doing at the Cyprus Police.