Description:
Since the 2009 NRC report [1], there has been a fundamental challenge in forensic science to establish a stronger scientific foundation and a statistical procedure for accurate firearm evidence identification and error rate estimation. To answer this challenge, researchers at NIST developed the Congruent Matching Cells (CMC) method [2,3], which is based on the principle of discretization that divides the entire image into cells, and uses subsequent cell correlations to quantify the topography similarity and pattern congruency of the correlated images. That makes it possible for ballistics identification and error rate estimation based on objective methods [3].
In addition to the congruent matching cells (CMC) method for correlation of breech face images, NIST researchers have recently developed the congruent matching cross-sections (CMX) method for firing pin correlations [4], the congruent matching profile segments (CMPS) method for bullet correlations [5], and the congruent matching features (CMF) method for a similarity map which shows the similar and dissimilar areas on the correlated images [6]. All validation tests for above methods using known matching (KM) and known non-matching images show clear separation between KM and KNM scores without any false identifications and false exclusions.
NIST has received requests from U.S. and international customers to provide a commercial system or source code based on the NIST invented congruent methods to support their firearm evidence identification and error rate estimation. The goal of the proposed SBIR project is an Automated System for Firearm Evidence Identification and Error Rate Estimation. This will be a commercialized system based on NIST invented CMC, CMX, CMPS and CMF methods. It will be used for automatic and objective firearm evidence identification covering all ballistics images including breech face, firing pin, ejector mark of cartridge cases and bullets. The output of the system will include an objective and conclusive result of identification or exclusion with false positive (for identifications) or false negative (for exclusions) error rate estimations. A similarity map is also developed from the system to visualize the similar and dissimilar areas on the correlated image pairs. Both the error rate estimation and the similarity mapping will provide a powerful tool to support ballistics examiners in court proceedings.
Phase I expected results:
• Based on the NIST invented CMC method [2,3], conduct a feasibility study for the development of a commercialized correlation program system using C++, OpenCV, Java, Python or other languages with high correlation speed and accuracy for breech face image correlations.
• Validation tests using two sets of breech face images in the NIST’s ballistics and toolmark research database, that include:
- Fadul dataset with 40 images including 63 KM and 717 KNM image pairs;
- Weller dataset with 95 images including 370 KM and 4095 KNM image pairs.
The correlation results must show clear separation between KM and KNM scores without any false identifications and false exclusions.
Phase II expected results:
Based on the NIST invented CMC [2,3], CMX [4], CMPS [5] and CMF [6] methods, develop an automated system for automatic and objective identification of all ballistics images including breech face, firing pin, ejector mark of cartridge cases and bullets.
• Validation tests for the developed commercial software using at least two sets of breech face, firing pin and ejector images of cartridge cases and at least two sets of bullet images in the NIST’s ballistics and toolmark research database. The correlation results must show clear separation between KM and KNM scores without any false identifications and false exclusions.
• A similarity map showing the similar and dissimilar areas in the correlated image pairs.
• A Statistical Fitting Program based on the NIST proposed statistical procedures [2,3] combined with an Error Rate Procedure [2,3] which can report the cumulative and individual error rates [2,3] for both the identification and exclusions conclusions using the CMC [2,3], CMX [4], CMPS [5] and CMF [6] methods for bullets and cartridge case correlations.
