Comparison of footwear retrieval systems for synthetic and real shoe marks

Shoe marks found on crime scenes can lead to the identification of the culprit, thus it is important to find the make and model of the footwear that left the marks. Some semi-automatic and automatic systems have been proposed in literature for the purpose, but they have been all tested on synthetic shoe marks, i.e. sole prints with added synthetic noise. Here we make a comparison of some of the methods reported in literature, both on synthetic and on real shoe marks coming from crime scenes: this last comparison has never been done before. Moreover we propose a new matching algorithm, based on the Mahalanobis distance, which is also compared to the other methods. Results show that simulated shoe marks are not suited to test a footwear retrieval system aimed at finding the shoe make and model of a shoe mark found on the crime scene.