An algorithm developed by a team of scientists from Malaysia attempts to clear up pictures for quantitative comparisons of firearms and toolmark evidence, as described in an open-access study published online today.

The analysis looked at 250 “noisy” images from five guns. The pistols were all the same make, model and caliber—Parabellum Vektor SP1 9mm— and all from a similar lot number manufactured in 2002, report the scientists in the Journal of Physics: Conference Series.

(The handguns are commonly in use among the Malaysian underworld, add the authors from several of the country’s universities.)

Their algorithm relied on three main components: image preprocessing, feature extraction and identification.

Key to the last two components are the mathematical representation of key traits: firing pin impression, breech face impression and the ejector impression. (These traits have all been key in automated identification systems already developed, including DRUGFIRE, IBIS, FIREBALL, the CIBLE system in France, and the TAIS system at use in Russia).

But for the first component—cleaning up a muddled image—they added “smoothing spatial filters” to clean up the “noise” in an image, they report.

Even when the noise is as high as 70 percent, their equation could automatically identify the subtle differences in the five very-similar pistols, they report.

“This study concludes that the proposed algorithm for firearm identification is robust when the images are contaminated with random-valued impulse noise with levels as high as 70 percent,” they write.

The work continues to consider further filters (switching median, multistate median, adaptive center filters) to remove further “noise” from the image comparisons.

Quantifying the comparison of firearms and toolmarks has been an objective of multiple teams of forensic scientists at various national and international agencies. For instance, the scientists at the National Institute of Standards and Technology in the U.S. has started work on a 3-D ballistics database which attempts to quantify match probabilities