Why Taliesin?

description psychology

Handwrittenand especially old documents aredifficult to read.To exploit them, we can use recognizers based onartificial intelligence..

backup storage

However, automatic recognition requiresa lot of training data.Therefore, thousands of examples need to be annotated.

find_in_page edit file_present

Taliesin facilitates theimport, slicingand annotationof handwritten documents and thus appears as a solution for the generation oflearning bases.A training base is a set of annotated examples containing images and their associated transcriptions.

Presentation of the project

Taliesin is a application for generating learning bases for handwriting recognition systems. These data allow recognizers to generate a model capable of making predictions on new documents.

To generate these learning databases, Taliesin offers an interface that facilitates the work of annotators. The training data sets are automatically generated thanks to deep neural network-based recognizers that annotate the different pages. In case of inconsistency, the user can modify the prediction manually using auto-completion. Once the image database is annotated, the user can export the examples and use them to train handwriting recognizers.

Interface

Home page

This page introduces Taliesin and allows the user to register or login.

Dashboard

Once connected, the user arrives on the dashboard. He can directly join the page that interests him:

Import
Cutting
Annotation
Administration
Parameters
Help

Import page

This page allows the user to import work bases and view them.

Cutting page

This page allows the user to draw the paragraphs when he does not have a file specifying them.
For this purpose, the drawing tools Rectangle and Polygon are presents.
Keyboard shortcuts are also available to facilitate annotation work.
For example, two keys are reserved for creating a new rectangle or polygon.
The Delete key is used to delete a shape.

Annotation page

This is the central page of the application. It allows the user to simply annotate the documents.
By selecting a phrase or through a keyboard shortcut, a window opens showing the user the predictions of the recognizers.
If these are not suitable, he can manually annotate the line with the help of auto-completion.
Various shortcuts have been added to speed up the annotation work. In particular, it is possible to move from one sentence to another, hide the results of the recognizers, validate the whole page and move to the next one...

Adminstration page

This page, reserved for administrators, allows to manage all the accounts. The roles of the users can be modified (annotator, project manager, administrator) as well as the email address. Accounts can also be deleted.

Our team

Our team is composed of seven students in their fourth year at the INSA Rennes INFORMATICS department.

Matisse BABONNEAU

Thomas BETTON

Corentin DUFOURG

Fabien LEFOYE

Elise MAUVIEUX

Glen POULIQUEN

Yuzhan WANG

Our partners

We would like to thank all our partners as well as our supervisors Alexandre GIMENEZ PUIG and ERWAN FOUCHE engineers at Sopra Steria as well as Bertrand COUASNON teacher researcher INSA/IRISA

Archives

The Archives départementales d'Ille-et-Vilaine provide us with handwritten documents and are part of the Taliesin beta testers.

Sopra Steria

French company of digital services. Two of its engineers bring us their experience to learn how to work better in a team, to manage a project in agile and to accompany us on the technical part.

INSA Rennes

Our engineering school thanks to which we were able to realize this project.

Doptim

Company creating AI and Big Data solutions in Brittany. One of its goals is to enable the transcription of old handwritten parish registers in the form of digital text. Doptim is one of Taliesin's beta-testers.

IntuiDoc

The IntuiDoc team of IRISA focuses its research on handwriting, gesture and document processing. The team provides us with recognizers and are beta-testers of Taliesin.