ReMatch: Re-identification of patterned species in open-set scenarios by matching keypoints and lines

Abstract This study presents ReMatch, a novel pipeline for individual re-identification of patterned species that combines keypoint and line matching within a probabilistic framework, enabling robust performance in open-set scenarios. Unlike many existing approaches that assume a fixed set of known individuals, ReMatch is designed to generalise effectively and detect previously unseen individuals without manual intervention. ReMatch has been evaluated across five diverse datasets (Balearic wall lizards, plains zebras, Saimaa ringed seals, whale sharks, and human fingerprints), demostrating a wide range of taxa and biometric modalities. Our method achieved superior performance compared to state-of-the-art deep learning models, reaching identification accuracies of 94.84% (closed-set) and 92.46% (open-set) in Balearic wall lizards. These results demonstrate ReMatch’s effectiveness even in the presence of deformations, viewpoint variation, resolution or limited training data. By integrating geometric feature extraction and probabilistic modelling, ReMatch offers a non-invasive, efficient, and broadly applicable solution for ecological monitoring, wildlife conservation, and biometric identification. Its ability to adapt across species with minimal retraining makes it particularly well suited for practical deployment in real-world field studies. Overall, ReMatch addresses key limitations of existing approaches and establishes a general-purpose tool for reliable, automated image-based identification.