Existing document parsing methods rely mainly on visual object detection techniques. They’re good at locating document elements, but face tough challenges in reconstructing hierarchical document structures, which greatly hinders the efficient conversion of unstructured text to structured knowledge. Traditional frameworks, constrained by predefined formats and inadequate multimodal integration (e.g., text, images and layout features), suffer from incomplete boundary detection and cross-page paragraph misidentification when processing complex document architectures or emerging document types. To address these challenges, this paper proposes the Extensible Document Parsing Structure Framework based on Font-Aware (EDPS-FA), which innovatively integrates Font-Aware semantic analysis with multimodal fusion to enhance parsing accuracy and robustness. The framework introduces three key innovations: (1) The Font-Aware Box Selection (FABS) that resolves layout parsing ambiguities, including in complete boundary recognition and accurate partition, through typographical feature extraction. (2) The Font-Aware Paragraph Extraction (FAPE) can accurately identify and analyze the cross-page paragraph information. (3) The Textual Tree JSON Reconstruction (TTJR) that extends conventional JSON-based structural representations by incorporating font-semantic relationships, enabling adaptive hierarchical reconstruction. At the same time, the optimal comparison enhancer (OCE) can optimize the analytical output through the multi-criteria alignment iteration method. Implemented with intelligent web-based parsing and batch processing capabilities, EDPS-FA demonstrates efficacy in processing structurally complex medical documents. Experimental validation through comparative and ablation studies achieved an F1-score of 90% across diverse parsing tasks, outperforming baseline methods in structural consistency. Available: Code is available at https://github.com/2439943021/EDPS-FA .