Approach to breech face impression comparison based on the robust estimation of a correspondence function

Area of Science:

• Forensic Science
• Ballistics
• Computer Vision

Background:

• Forensic firearm analysis relies on matching tool-marks from ammunition to specific firearms.
• Automated comparison of breech face impressions using Scale Invariant Feature Transform (SIFT) and RANdom SAmple Consensus (RANSAC) has been proposed.
• Existing methods require improvement in robustness and repeatability for reliable feature matching.

Purpose of the Study:

• To propose an estimation method for establishing a correspondence function between features of comparison impression pairs.
• To enhance the robustness and repeatability of automated feature matching in forensic firearm analysis.
• To improve the accuracy of identifying firearm-ammunition associations.

Main Methods:

• An iterative algorithm employing Support Vector Regression (SVR) to estimate the correspondence function.
• A robust weighting method to exclude outliers among putative correspondences.
• Consistency detection to mitigate the over-fitting problem in SVR.
• Validation on three distinct datasets of cartridge case breech face impressions (Fadul, Weller, Lightstone).

Main Results:

• The proposed method successfully distinguished between known matching (KM) and known non-matching (KNM) impression pairs.
• KM pairs consistently showed over 20 matching feature points, while KNM pairs had 3-8 correspondences.
• The method demonstrated advantages for granular impressions but had limitations with striation marks.
• Retained more reliable matching feature points compared to RANSAC.

Conclusions:

• The proposed estimation method is feasible for ballistic feature comparison.
• It significantly improves the repeatability of feature correspondence selection in automated firearm analysis.
• The technique offers a more robust approach to forensic firearm identification, particularly for specific impression types.