Rejoining oracle bone fragments plays an import role in studying the history and culture of the Shang dynasty by its characters. However, current computer vision technology has a low accuracy in judging whether the texture of oracle bone fragment image pairs can be put back together. When rejoining fragment images, the coordinate sequence and texture features of edge pixels from original and target fragment images form a continuous symmetrical structure, so we put forward an internal similarity network (ISN) to rejoin the fragment image automatically. Firstly, an edge equidistant matching (EEM) algorithm was given to search similar coordinate sequences of edge segment pairs on the fragment image contours and to locally match the edge coordinate sequence of an oracle bone fragment image. Then, a target mask-based method was designed in order to put two images into a whole and to cut a local region image by the local matching edge. Next, we calculated a convolution feature gradient map (CFGM) of the local region image texture, and an internal similarity pooling (ISP) layer was proposed to compute the internal similarity of the convolution feature gradient map. Finally, ISN was contributed in order to evaluate a similarity score of a local region image texture and to determine whether two fragment images are a coherent whole. The experiments show that the correct judgement probability of ISN is higher than 90% in actual rejoining work and that our method searched 37 pairs of correctly rejoined oracle bone fragment images that have not been discovered by archaeologists.

In recent years, high-fidelity three-dimensional (3D) oracle bone models (3D-OBMs) have received extensive attention from oracle bone experts due to their unparalleled reducibility to real oracle bone. In the research process of 3D-OBMs, the first procedure is to extract oracle bone inscriptions (OBIs) from the model to form individual oracle bone characters (OBCs). However, the manual extraction of OBIs is a time-consuming and labor-intensive task that relies heavily on oracle bone knowledge. To address these problems, we propose a texture-mapping-based OBI extractor (tm-OBIE), which leverages the symmetrical characteristics of the texture mapping process and is able to extract 3D-OBIs from 3D-OBMs saved as a wavefront file. The OBIs in the texture file were first located using a trained 2D object detector. After that, the 3D mesh area, where the OBIs are located, was obtained using an inverse texture mapping method. Thirdly, a specific 2D plane was fitted using the centroid of triangular faces in the flat regions of the mesh via a singular value decomposition (SVD) method. Finally, by measuring the distances between the triangle meshes and the fitted plane, the meshes of the 3D-OBIs were obtained. This paper verifies the feasibility of this method via experiments and analyzes the possibility of using the algorithm framework for extracting other ancient characters from their corresponding 3D models.

Oracle bone inscriptions, the earliest mature Chinese writing system, are a primary source for studying ancient Chinese civilization, yet their interpretation remains labor-intensive. Automatically detecting and recognizing visible characters on rubbing images is therefore an entrance task for AI-assisted interpretation. The Oracle Bone Inscription Multi-modal Dataset exposes the problem of extreme long-tailed class distribution: aggregate evaluation conceals severe performance defects on rare classes—the detector can reach an overall AP₅₀ of 0.63, while the few-shot class AP₅₀ is only 0.06. We conduct a systematic diagnosis with a frequency-aware stratified evaluation protocol. Controlled comparisons show that localization is less limiting than rare-glyph recognition, and that in-head and crop-level remedies trade aggregate accuracy against tail accuracy. We thus propose FAR-YOLO, a minimal decoupled baseline that separates localization and recognition training, optimizing the recognition branch only with frequency-aware sampling and morphology-preserving augmentation. While maintaining nearly unchanged overall AP₅₀, FAR-YOLO improves the few-shot AP₅₀ by 10.5 percentage points. It ranks second on all four frequency-stratified metrics among representative long-tailed methods, with consistent tail gains across seven recognition backbones.