Multi modal information extraction for visually rich documents

The main objective of this project is to extract information from documents (Name, issue date, address, account number, etc.) that have diverse set of layouts (bank statements, utility bills, credit card statements, etc.). This problem is challenging because of diversity in layouts, variety of languages and complexity of template structures, which makes traditional NLP approaches difficult to use. The goal is to develop a solution that can make use of not only textual but visual layout information to extract information. Intern will be involved in reading research paper pertaining the problem, analysing image datasets and write code to run experiments on public and proprietary datasets to improve the accuracy of the system. This project is part of Jumio’s Document Verification product, that provides companies with services that verifies customer information through the analysis of their documents.

Faculty Supervisor:

Ioannis Mitliagkas

Student:

Partner:

Jumio

Discipline:

Computer science

Sector:

Professional, scientific and technical services

University:

Université de Montréal

Program: