Automated comparison of firearm bullets.

Fired bullets bear striation marks that can be thought of as a "fingerprint" left by the firearm. This new comparison approach is based on an automated extraction of a "signature" encompassing the relevant marks from an image. To this end, multiple pictures of the bullet are recorded first by using different illumination patterns, and a high quality image is generated by means of fusion techniques. After a preprocessing, the image intensities are filtered along the striations direction, yielding a compact representation of the marks. A non-linear filter selects the striae of interest. The actual comparison takes place by cross-correlating the signatures obtained this way. Finally, an assessment strategy is proposed to objectively evaluate the performance of the system. As demonstrated with an image database of real bullets, the proposed approach outperforms a state-of-the-art commercial system.