Towards Automatic Computation of the SkullCode: A Novel Biometric Approach for Cranial Identification

One of the main activities of forensic anthropologists consists of identifying living or deceased individuals from the analysis of physiological characteristics. In this context, recent studies evaluated the possibility of using novel physiological traits. Furthermore, recent works aim at realizing decision support tools for forensic scientists based on artificial intelligence techniques. In our previous work, computer scientists and forensic anthropologists collaborated to design a novel biometric recognition approach based on non-metric cranial traits, resulting in a 35-digit representation of these features (referred to as SkullCode). In this paper, we take a step towards the automated detection of SkullCode features. Specifically, we explore the possibility of classifying one of its traits by using deep neural networks. Our approach involves the extraction of two-dimensional slices from the skull’s region of interest, followed by a deep neural network that classifies the presence or absence of the trait. To validate our method, in collaboration with Italian hospitals, we created a dataset of 952 annotated samples from 84 individuals. The best obtained accuracy is equal to 88.87%. This promising result highlights the potential of our method for real-world applications and bring us closer to developing a fully automated identification system based on SkullCode features.