Recently, an article in Evidence Magazine discussed how hot and cold pixels in a camera can be used to fingerprint images from that camera, and thereby convict a suspect based on pictures and camera equipment, one or more of which is found in the suspects’ possession.
This is a clear indicator of why the defense needs quality representation and expert witnesses, and State appointed attorneys may miss crucial arguments without proper expert representation.
The article makes it seem quite easy to match camera to image, but it omits a couple of possibilities that drastically complicate the process and likelihood of conviction. Here are two additional complications that can ruin a case, and there are likely more; each situation is unique.
First, are the criminal images captured compressed or uncompressed? While uncompressed images are used for validating the hot and cold pixel fingerprint of the camera, they can only be matched to illegal images that are likewise uncompressed, or mathematically validated if compressed. Video cameras are likely to use compression to store the images, and photographs are frequently taken using image compression or scene magnification, any of which must be accounted for when eliminating possible errors in verification of the images and camera.
Worse, the author expresses probability of error in terms that appear astronomical, using math to paint a rosy picture of probability that fails to account for additional possibilities that must be taken into account before convicting based on pixel fingerprint evidence.
An example of such an assumption is that those pixels are unique to that camera, when in fact, hot or cold pixels can be endemic to the entire product line of cameras or image sensors. Two ways manufacturing can spoil the odds is by introducing defects to the image sensors when they are being mounted in the camera, or by damaging the sensors in manufacturing of the silicon wafers.
If a production line introduced hot or cold pixels, before going to trial we need to know what the manufacturers criteria is for acceptable bad pixels, and we need to know the production statistics. In the example given, any or all of the four hot pixels could have been present since the camera was made, which skews the probability projection. If all four were manufacturers defects, or cannot be shown to not be manufacturers defects, the fingerprinting process proves nothing, not even the camera family, as imaging chips could be used in multiple camera lines.
Other factors such as the number of cameras in the geographic area complicate or simplify fingerprinting pixels for matching. For instance, if the pictures and camera are found on the person in the Outback, miles from anyone else, the probability of responsibility rises to near positive assumption. But a common camera taken in New York may mean an uphill battle to reach a probability acceptable to a jury.
Improper representation of evidence leads to false convictions of innocent people. Make sure you do it 100% right, and any argument using statistics and probability needs to be examined closely to locate additional factors not taken into account.
This blog and it’s contents copyright 2009 Digital Trust, LLC. Republication of this post is permitted provided the Digital Trust name, url, and this paragraph are included. Counsel, do not list without contract.