Source Identification of 3D Printer Based on Layered Texture Encoders

Abstract

With the rapid growth in the three-dimensional (3D) printing content market, various unprecedented criminal cases and copyright protection issues have emerged. In response to this imminent and emergent difficulty, we propose a forensic technique for identifying the source of 3D printed products based only on surface inspection features. The surface texture of 3D printed objects exhibits, inevitably, extremely fine periodic features during the additive manufacturing process. We propose a two-stream texture encoder, referred to as CFTNet, combined with fast Fourier transform and positional encoding of the transformer encoder to leverage inherent periodic features occurring during the additive manufacturing. As benchmarks, we define detailed scenarios for six source identification problems and present detailed verification procedures with a large-scale benchmark dataset SI3DP++ for forensic real-world scenarios. A certain level of performance was achieved using six benchmarks, including printer and device-level identification. Moreover, we extended the baseline study based on the benchmark set to forensic test scenarios from multiple perspectives in preparation for real situations. We reveal both the dataset and detailed experimental design to provide an opportunity to facilitate future in-depth studies related to forensics and protection of intellectual property.