Identification of fossil hominin cranial fragments: a computer vision-based approach

International audience Cranial fragments are among the most commonly recovered hominin remains in the fossil record (e.g., [1]). Despite their potential to contribute to taxonomic discussions [2-3], these fragments are often difficult to anatomically assign, especially when originating from the cranial vault, and this is particularly true for Plio-Pleistocene hominin assemblages. As a result, isolated cranial fragments that cannot be confidently identified are frequently excluded from analyses, leading to the loss of valuable data on early hominin cranial anatomy. Improving our ability to determine the anatomical location of these isolated remains would not only increase the number of specimens available for study, but also enhance our understanding of the evolutionary processes (whether neutral or adaptive) having shaped the hominin cranium and facilitate more detailed reconstructions of hominin brain changes through the description of endocranial imprints. Within this context, our project aims to develop a computer vision-based approach for automatically identifying the anatomical position of isolated fossil hominin cranial remains based on a number of predefined textural, structural and geometrical criteria.Our image-based approach relies on the quantitative analysis of bone properties across complete extant human crania from the Pretoria Bone Collection of the University of Pretoria (South Africa) and from PALEVOPRIM (France) which were scanned using micro-tomography at the South African Nuclear Energy Corporation (Necsa) and at the PLATINA (IC2MP) platform in Poitiers, respectively, and served as references. We focused on identifying and prioritizing a set of diagnostic features described in the literature as informative for cranial bone identification, that includes the presence and morphology of cranial sutures, bone curvature, bone thickness, structural arrangement of the diploic layer and inner and outer tables, as well as bone density [2-4]. Our approach, partly based on existing literature combining 3D data and computer-assisted methods (e.g., [5]), consists in extracting different types of gradients (e.g., curvature, density, thickness) associated with specific locations in the cranium by comparison with a reference model. We applied this approach to hominin cranial fragments recovered from the site of Sterkfontein and scanned at the Evolutionary Studies Institute of the University of the Witwatersrand (South Africa).Our approach offers a reliable framework for testing hypotheses about the anatomical identification of isolated cranial fragments, as well as for generating informed proposals regarding the possible anatomical origin of undetermined specimens within the hominin fossil record. Moreover, this work constitutes a foundational step toward more comprehensive studies integrating the morphological complexity of the hominin cranium and, ultimately, proposing tentative reconstructions of entire crania from fragmentary remains, thereby expanding the available dataset for investigating cranial evolution in early hominins.