Blueprint

Blueprint document automatic recognition using deep learning

(TeamUp 2019 (Innoviris))

Description:

Symbol recognition, and in particular optical character recognition (OCR), can be consider as a mature domain. However when it comes to structured document, in particular schematics or blueprints, the complexity and the specificity of the document is increasing dramatically. The usual approach relies on line identification, symbol recognition and more generally OCR limited to portion of the document. Often ontologies, that are machine-compatible description of concepts, are defined to describe/analyse the document in a structured way. Ontologies are domain specific and require an important expert input. Blueprint project proposes to implement and train state of the art techniques of machine vision in order to provide the user with an interactive constantly learning system that will reinforce its prediction accuracy by a continuous human interaction. Moreover the chosen approach will enable to apply the same framework to dataset from diverse origin, in order to cope with the versatility of the blueprint digitization problem.

Partners:

A.K.K.A. consulting group

Responsible:

  • DEBEIR Olivier

Members:

  • HAHN Stéphan

Support from:

  • Région bruxelloise