Mélodie Boillet's thesis defence on 10 January at 2pm

Date :


Mélodie Boillet will defend her thesis on 10 January on Tuesday, January 10th at 2pm at the Sophie Germain amphitheater (D) of the UFR Sciences et Techniques, at the Madrillet campus (Saint-Étienne-du-Rouvray).

This thesis was carried out in the context of a CIFRE collaboration between TEKLIA and the LITIS Apprentissage team (Rouen-Normandy University) and is entitled:

"Object Detection in Digitized Documents by Neural Networks".

The defense will take place in front of a jury composed of:

  • Andreas Fischer, Professor at the Institute of Artificial Intelligence and Complex Systems in Fribourg (Switzerland), Referee
  • Harold Mouchère, Professor at the University of Nantes, Referee
  • Laurence Likforman, Associate professor at Télécom ParisTech, Examiner
  • Caroline Petitjean, Professor at the University of Rouen, Examiner
  • Thierry Paquet, Professor at the University of Rouen, Thesis director
  • Christopher Kermorvant, President of Teklia, Thesis supervisor

The public defense will take place in a hybrid format. Those who wish to attend the defense remotely are invited via the following Zoom link:

Participer à la réunion Zoom

ID de réunion : 943 5025 9642
Code secret : R7YW94


Documents constitute a valuable collection of information that is usually difficult to access. The transformation of these documents into digital documents is now possible through their digitization and the automatic extraction of their contents. This extraction requires the detection of different elements such as text lines, which are essential to obtain the transcription of the image's textual contents.

In this thesis, we study different tasks related to document layout analysis such as the detection of text lines, the splitting into acts or the detection of the writing support. Our objective is to propose models allowing the detection of objects while considering the difficulties associated with document processing and the constraints related to their use in an industrial context. We will propose a complete study of the object detection task in document images by analyzing annotations, different data selection strategies and evaluation protocols.