A Survey on Reading Order, Table of Contents, and Structure Extraction in Document Analysis

Simone Giovannini, Simone Marinai; Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops, 2025, pp. 7585-7594

Understanding the structure of complex documents is a challenging goal in document analysis, encompassing tasks such as reading order identification, table of contents extraction, and hierarchical structure prediction. This survey has a twofold aim: first to categorize these tasks based on their problem formulation, identifying key challenges and defining the scope of each task. Second, to group the proposed methods for addressing these tasks according to the fundamental components they rely on. We analyze their methodologies, advantages, and limitations across various document types, highlighting recent advances in deep learning and multimodal processing. Additionally, we present comparative insights into different paradigms, such as graph neural networks and attention-based models, to identify promising directions for research in this area. Although we do not include a quantitative comparison of methods or datasets due to space constraints, this work serves as a first step toward a more comprehensive evaluation, aiming to highlight conceptual similarities and differences that can guide future research.