Estimating probabilities in forensics is a complex problem. There are several methodologies, all facing difficulties from the number of variables and our lack of knowledge of the distribution of some of the variables.

Tang and Srihari review prior methods and derive a way of estimating the likelihood ratio that is both mathematically sound and able to be explained to non-mathematicians, although this paper is quite mathematical. They evaluate several methods using actual data from handwriting and shoe prints. Their solution involves the product of a distance metric and a rarity metric.

Not only does emphasizing unusual features in the calculation improve performance in realistic cases, but it conforms to naive human expectations about what should matter. This is a significant paper in a steadily more important area.