Daniel@FinTOC-2021: Taking Advantage of Images and Vectorial Shapes in Native PDF Document Analysis

In this paper, we present our contribution to the FinTOC-2021 Shared Task “Financial Document Structure Extraction”. We participated in the tracks dedicated to English and French document processing. We get results for Title detection and TOC generation performance which demonstrates a good precision. We address the problem in a fairly unusual but ambitious way which consists in considering simultaneously text content, vectorial shapes and images embedded in the native PDF document, and to structure the document in its entirety.