On Graph Representation based Re-Identification A Proof of Concept

This submission introduces an innovative method based on graph representation for re-identifying chipwood surface structures. In the specific application under consideration, it also addresses the re-identification of Euro-pallets. To achieve this objective, we propose a departure from conventional reidentification methods by advocating the substitution of the standard image representation with a significantly compressed graph representation. This not only facilitates the development of an efficient algorithm but also imparts resilience to environmental variations, including rotation and shearing. The resultant approach, named "IRAG" (Image Representation through Anomaly Graphs), manifests as a siamese graph neural network. This network is employed on a pre-existing dataset, which comprises images sourced from 502 EPAL pallet blocks and has been previously published. The obtained results from this approach show a rank-1 accuracy of 27 Additionally, the experiments conducted in this work demonstrate that the re-identification accuracy of the model is not affected by rotation or shearing, indicating the model's invariance to these environmental changes. This paper represents an updated and improved version of the paper "Towards graph representation-based re-identification of chipwood pallet blocks" published at ICMLA 2022.