A CBAM-Enhanced PYGT-CRNN-CTC Framework for Recognizing Complex Traditional Chinese Characters on Stone Inscriptions

View at publisher Summarise with VIRAT
Authors: Shiyi Tan, Juan Xv, Mingjun Qiu, Junling Zhu, Jie Li
Cited by: 0
References: 6

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Stone steles and epigraphic inscriptions are vital carriers of Chinese cultural, historical, and linguistic heritage. However, automatic recognition of Traditional Chinese characters on stone inscriptions remains highly challenging due to erosion, uneven carving, illumination variation, biological contamination, and severe structural distortions such as stroke breakage and deformation. Existing Optical Character Recognition (OCR) systems, primarily designed for modern documents, rely on grid-based feature representations and therefore struggle to model the non-Euclidean structural characteristics of ancient scripts. To address these challenges, this study proposes an end-to-end OCR framework tailored for degraded stone inscriptions. The core innovation is the Pyramid Graph Transformer (PYGT), which represents characters as graphs and enables unified global, local, and multi-scale structural modeling. Specifically, PYGT integrates graph attention to capture long-range dependencies, graph convolution to model local stroke connectivity, and a pyramid-style graph coarsening strategy to learn hierarchical multi-resolution features. This graph-based abstraction fundamentally distinguishes PYGT from conventional Convolutional Neural Networks (CNNs) and flat Transformer architectures. Combined with a CNN with Convolutional Block Attention Module (CNN–CBAM), a bidirectional Gated Recurrent Unit (GRU) encoder, and a Connectionist Temporal Classification (CTC) decoder, the proposed framework achieves superior recognition performance.

Fields of Study & Subjects

Handwritten Text Recognition Techniques Vehicle License Plate Recognition Advanced Neural Network Applications

Publication details

Journal / source
International Conference on Artificial Intelligence and Social Network Systems (AISNS)
Pages
92–95
Publication date
9 January 2026
Language
EN