Physical Matching Verification
Yoram Yekutieli, Yaron Shor, Sarena Wiesner, Tsadok Tsach
See also the instructional videos on "Tool Mark Evidence"
Executive summary
We developed a prototype system used for physical matching in 2D. The system has two main functions: one is to assist forensic experts in performing physical matching in an objective manner, and the second to collect statistics and build confidence levels regarding the physical matches.
The first function receives input images of broken or torn pieces photographed on distinctive background. The system then extracts the contour of each piece and lets the user choose a segment on one of the contours as a target for matching. The system then runs a search on the other contours, finds potential matches and orders these candidates according to the quality of their fit with the target. Using the statistics from the second function the user can estimate the potential error rates of each candidate matching pair.
The second function takes as an input multiple images of broken or torn pieces and then generates many examples of physical matching that are classified into two populations: the correct matching pairs and the non-matching pairs. The system gathers the frequencies of the matching error for the two populations and presents the results as two histograms. The outputs of the system are several statistical measures describing the quality of matching as a function of the matching error value. The user can find the probabilities of a match being right or wrong for any specific value of matching error and thus is able to compare between matching pairs, and find out which is the best evidence, even among different sizes or materials.
We used the second function to collect statistics for different fracture line lengths of three different materials: silicon, metal coated paper and Perspex. The total error rates 4 (false negative + false positive) for 1 cm matches were 0.007 for the silicon, 0.37 for the paper and 0.4 for the Perspex.
The statistical results were much lower than initially expected. This is because we used only the 2D fracture lines and not any additional clue commonly used in toolmark comparison such as the 3D nature of some fractures or any existing texture and graphic patterns on the surface or outer border of the pieces to be compared. We also classified the pairs into matches and non-matches and ignored a third classification – “inconclusive”. For all these reasons, our results are the entering point to the numerical or quantitative evaluation. Adding the surface details or performing a 3D match, can bring us much closer to the results achieved by toolmark experts.
In presenting results of scientific nature to court, the law system in many countries, demands having a potential error rate for the results. Until today, there wasn't any numerical way to calculate the potential or known error rate, for 2D or 3D comparisons. This project supplies the forensic scientist the essential tools to calculate this crucial information.
The results of this research, experts can express their findings in a more quantitative way. It is in the reach of every scientist, holding this computerized statistics generator, to create a new database for any material. Computing the potential error rate in physical match comparison for objects received in the laboratory becomes possible.