Tutorial 1: Beyond Recognition: A Multidimensional Exploration of Characters

Organizer: Seiichi Uchida (Kyushu University, Japan), uchida@ait.kyushu-u.ac.jp

In the field of document image recognition and understanding, characters are an indispensable element. Traditionally, much research has focused primarily on recognizing characters — treating them simply as objects of recognition — and on extracting semantic information from text when integrating with natural language processing. However, we seldom pause to ponder a fundamental question: What exactly is a character?

This tutorial aims to investigate that question from multiple perspectives, shedding light on how character images differ from generic images, how fonts are used and why so many different fonts are necessary. We will also explore why we are able to identify certain shapes as specific characters (e.g., recognizing an “A” as the letter A) and examine the unique properties that distinguish handwritten text from printed text. Moreover, we will discuss how characters appear in various contexts — from printed documents to building surfaces and inscriptions — and the cultural and historical dimensions surrounding them.

By exploring these facets, participants will gain fresh insights into the fundamental nature of textual information. This broader perspective may lead to innovative approaches in document image analysis and open new avenues for research and practical applications within the ICDAR community. Rather than focusing on incremental improvements to existing tasks, I hope this tutorial will inspire entirely new lines of research on the nature of characters!

Tutorial 2: How to train a Multi-modal Large Document Understanding Model?

Organizers:

Wei Shen (Shanghai Jiao Tong University), wei.shen@sjtu.edu.cn

Yu Zhou (Nankai University), yzhou@nankai.edu.cn

This tutorial will provide participants with a comprehensive understanding of multi-modal document understanding models, focusing on the integration of textual and visual data for improved document analysis.

Attendees will learn about model architecture, training techniques, and real-world applications. The tutorial is designed for researchers and practitioners interested in the latest advancements in document analysis, making it a perfect addition to the ICDAR conference program.

Tutorial 3: General Introduction to Oracle Bone Scripts Processing

Organizers:

Qiufeng Wang (Xi’an Jiaotong-Liverpool University), Qiufeng.Wang@xjtlu.edu.cn

Yuliang Liu (Huazhong University of Science and Technology), ylliu@hust.edu.cn

Bang Li (Anyang Normal University), libang@aynu.edu.cn

Oracle Bone Script (OBS) is the earliest known form of Chinese writing, dating back over 3,000 years to the Shang Dynasty. It serves as a crucial resource for historical linguistics, epigraphy, and the study of ancient Chinese civilization. However, due to the complexity of OBS, including its vast character variations, irregular inscriptions, and fragmented artifacts, the automatic recognition, segmentation, and analysis of OBS remain highly challenging tasks in document analysis and recognition.

This tutorial aims to introduce participants to the latest advances in Oracle Bone Script processing, covering key techniques such as character recognition, inscription restoration, dataset construction, and deep learning-based approaches. We will also discuss the interdisciplinary collaboration between computer vision, natural language processing, and archaeology, which is essential for advancing research in this domain.

The motivation behind this tutorial is twofold: First, despite significant progress in optical character recognition (OCR) for modern scripts, ancient scripts like OBS require specialized methods due to their non-standard writing systems. Second, there is growing interest in leveraging artificial intelligence (AI) and deep learning for historical document analysis, making this an opportune moment to bring together researchers from diverse backgrounds to explore innovative solutions.

By providing a comprehensive overview of OBS processing, this tutorial will foster collaboration between researchers in document analysis, computational humanities, and epigraphy, ultimately contributing to the preservation and understanding of one of the world's most ancient writing systems.