Publications · Teklia

Vezina, Helène; Bournival, Jean-Sebastien; Kermorvant, Christopher; Bonhomme, Marie-Laurence
I-Balsac: Completing Families with the Help of Automatic Text Recognition conference
Annual meeting of the Social Science History Association , 2019

Boros, Emanuela; Toumi, Alexis; Rouchet, Erwan; Abadie, Bastien; Stutzmann, Dominique; Kermorvant, Christopher
Automatic page classification in a large collection of manuscripts based on the International Image Interoperability Framework inproceedings
International Conference on Document Analysis and Recognition , 2019

Boillet, Mélodie; Bonhomme, Marie-Laurence; Stutzmann, Dominique; Kermorvant, Christopher
HORAE: an annotated dataset of books of hours inproceedings
International Workshop on Historical Document Imaging and Processing , 2019

Stutzmann, Dominique; Currie, Jacob; Daille, Béatrice; Hazem, Amir; Kermorvant, Christopher
Integrated DH. Rationale of the HORAE Research Project conference
Digital Humanities , 2019

Renton, Guillaume; Soullard, Yann; Chatelain, Clément; Adam, Sébastien; Kermorvant, Christopher; Paquet, Thierry
Fully Convolutional Network with dilated convolutions for Handwritten text line segmentation article
International Journal on Document Analysis and Recognition , 2018

Moysset, Bastien; Kermorvant, Christopher; Wolf, Christian
Learning to detect, localize and recognize many text objects in document images from few examples article
International Journal on Document Analysis and Recognition , 2018

Bluche, Théodore; Kermorvant, Christopher; Touzet, Claude; Glotin, Hervé
Cortical-Inspired Open-Bigram Representation for Handwritten Word Recognition inproceedings
International Conference on Document Analysis and Recognition , 2017

Bluche, Théodore; Hamel, Sebastien; Kermorvant, Christopher; Puigcerver, Joan; Stutzmann, Dominique; Toselli, Alejandro; Vidal, Enrique
Preparatory KWS Experiments for Large-Scale Indexing of a Vast Medieval Manuscript Collection in the HIMANIS Project inproceedings
International Conference on Document Analysis and Recognition , 2017

Moysset, Bastien; Kermorvant, Christopher; Wolf, Christian
Full-Page Text Recognition: Learning Where to Start and When to Stop inproceedings
International Conference on Document Analysis and Recognition , 2017

Renton, Guillaume; Chatelain, Clément; Adam, Sébastien; Kermorvant, Christopher; Paquet, Thierry
Handwritten text line segmentation using Fully Convolutional Network inproceedings
International Workshop on Historical Document Imaging and Processing , 2017

Bluche, Théodore; Kermorvant, Christopher; Ney, Hermann
How to Design Deep Neural Networks for Handwriting Recognition inbook
Pirlo, Byron Leite Dantas Bezerra Cleber Zanchettin Alejandro H. Toselli Giuseppe , Nova Science Pub Inc , 2017

Moysset, Bastien; Louradour, Jérôme; Kermorvant, Christopher; Wolf, Christian
Learning text-line localization with shared and local regression neural networks inproceedings
International Conference on Frontiers in Handwriting Recognition , 2016

Bluche, Théodore; Stutzmann, Dominique; Kermorvant, Christopher
Automatic Handwritten Character Segmentation for Paleographical Character Shape Analysis inproceedings
Document Analysis Systems , 2016

Abstract

Written texts are both physical (signs, shapes and graphical systems) and abstract objects (ideas), whose meanings and social connotations evolve through time. To study this dual nature of texts, palaeographers need to analyse large scale corpora at the finest granularity, such as character shape. This goal can only be reached through an automatic segmentation process. In this paper, we present a method, based on Handwritten Text Recognition, to automatically align images of digitized manuscripts with texts from scholarly editions, at the levels of page, column, line, word, and character. It has been successfully applied to two datasets of medieval manuscripts, which are now almost fully segmented at character level. The quality of the word and character segmentations are evaluated and further palaeographical analysis are presented.

Bluche, Théodore; Kermorvant, Christopher; Ney, Hermann; Louradour, Jerome
La CTC et son intrigant label inproceedings
Colloque International Francophone sur l’Écrit et le Document 2016 (CIFED) , 2016

Abstract

Les systèmes de reconnaissance d’écriture vainqueurs d’évaluations internationales ces dernières années sont basés sur des réseaux de type LSTM, entraînés avec un critère de classification temporelle connexionniste (CTC). L’algorithme de la CTC est basé sur une procédure "forward-backward", sans segmentation de la séquence d’entrée avant l’entraînement. Les sorties du réseau sont les caractères à modéliser, auxquels on ajoute un label spécial, noncaractère, (blank). D’autre part, dans les systèmes hybrides réseaux de neurones / modèles de Markov cachés (MMCs), les réseaux sont entraînés au niveau trame à prédire des états de MMC. Dans cet article, nous montrons que la CTC est une forme dérivée de l’entraînement forward-backward de MMCs, et qu’elle peut donc être étendue à des topologies arbitraires de MMC. Nous appliquons cette méthode à des perceptrons multicouches et la comparons à l’entraînement au niveau trame dans diverses situations. Enfin, nous analysons le rôle intrigant de ce label spécial "blank".

Stutzmann, Dominique; Bluche, Théodore; Kermorvant, Christopher; Englin, Véronique; Vincent, Nicole; Leydier, Yann; Cloppet, Florence
Text-Image Alignment and Automated Letter-form Classification: Reading vs. Looking at inproceedings
Second International Conference on Natural Sciences and Technology in Manuscript Analysis , 2016

Moysset, Bastien; Kermorvant, Christopher; Wolf, Christian; Louradour, Jérôme
Paragraph text segmentation into lines with Recurrent Neural Networks inproceedings
International Conference of Document Analysis and Recognition , 2015

Stutzmann, Dominique; Bluche, Théodore; Lavrentev, Alexei; Leydier, Yann; Kermorvant, Christopher
From Text and Image to Historical Resource : Text-Image Alignment for Digital Humanists inproceedings
Digital Humanities , 2015

Bluche, Théodore; Ney, Hermann; Kermorvant, Christopher
Framewise and CTC Training of Neural Networks for Handwriting Recognition inproceedings
International Conference of Document Analysis and Recognition , 2015

Bluche, Théodore; Ney, Hermann; Kermorvant, Christopher
The LIMSI Handwriting Recognition System for the HTRtS 2014 Contest inproceedings
International Conference of Document Analysis and Recognition , 2015

Benzeghiba, Mohamed Faouzi; Louradour, Jérôme; Kermorvant, Christopher
Hybrid Word / Part-of-Arabic-Word Language Models for Arabic Text Document Recognition inproceedings
International Conference of Document Analysis and Recognition , 2015

Bluche, Théodore; Kermorvant, Christopher; Louradour, Jérôme
Where to Apply Dropout in Recurrent Neural Networks for Handwriting Recognition ? inproceedings
International Conference of Document Analysis and Recognition , 2015

Links (2)

Moysset, Bastien; Messina, Ronaldo; Kermorvant, Christopher
A Comparison of Recognition Strategies for Printed / Handwritten Composite Documents inproceedings
International Conference on Frontiers in Handwriting Recognition , 2014

Louradour, Jérome; Kermorvant, Christopher
Curriculum Learning for Handwritten Text Line Recognition inproceedings
Document Analysis Systems , 2014

Pham, Vu; Bluche, Théodore; Kermorvant, Christopher; Louradour, Jérôme
Dropout improves recurrent neural networks for handwriting recognition inproceedings
International Conference on Frontiers in Handwriting Recognition , 2014

Moysset, Bastien; Bluche, Théodore; Knibbe, Maxime; Benzeghiba, Mohamed Faouzi; Messina, Ronaldo; Louradour, Jérôme; Kermorvant, Christopher
The A2iA Multi-lingual Text Recognition System at the Maurdor Evaluation inproceedings
International Conference on Frontiers in Handwriting Recognition , 2014

Bluche, Théodore; Louradour, Jérôme; Knibbe, Maxime; Moysset, Bastien; Benzeghiba, Mohamed Faouzi; Kermorvant, Christopher
The A2iA Handwritten Arabic Text Recognition System at the OpenHaRT2013 Evaluation Campaign inproceedings
Document Analysis Systems , 2014

Messina, Ronaldo; Kermorvant, Christopher
Surgenerative Finite State Transducer n-gram for Out-Of-Vocabulary Word Recognition inproceedings
Document Analysis Systems , 2014

Bluche, Théodore; Moysset, Bastien; Kermorvant, Christopher
Automatic Line Segmentation and Ground-Truth Alignment of Handwritten Documents inproceedings
International Conference on Frontiers in Handwriting Recognition. , 2014

Touzet, Claude; Kermorvant, Christopher; Glotin, Hervé
A Biologically Plausible SOM Representation of the Orthographic Form of 50,000 French Words inproceedings
Workshop on Self-Organizing Maps. , 2014

Bluche, Théodore; Ney, Hermann; Kermorvant, Christopher
A Comparison of Sequence-Trained Deep Neural Networks and Recurrent Neural Networks Optical Modeling for Handwriting Recognition inproceedings
International Conference on Statistical Language and Speech Processing , 2014

Kermorvant, Hervé Glotin Xanadu Halkias Christopher
Evolution of Sparse Constraints in Deep AI: From a general sparse penalty to Structured Sparsity techreport
2013

Nion, Thibauld; Menasri, Farès; Louradour, Jérôme; Sibade, Cédric; Retornaz, Thomas; Métaireau, Pierre-Yves; Kermorvant, Christopher
Handwritten information extraction from historical census documents inproceedings
International Conference of Document Analysis and Recognition , 2013

Bluche, Théodore; Ney, Hermann; Kermorvant, Christopher
Tandem HMM with convolutional neural network for handwritten word recognition inproceedings
International Conference on Acoustics Speech and Signal Processing , 2013

Abstract

In this paper, we investigate the association of hidden Markov models and convolutional neural networks for handwritten word recognition. The convolutional neural networks have been successfully applied to various computer vision tasks, including handwritten character recognition. In this work, we show that they can replace Gaussian mixtures to compute emission probabilities in hidden Markov models (hybrid association), or serve as feature extractor for a standard Gaussian HMM system. The proposed systems outperform a basic HMM based on either decorrelated pixels or handcrafted features. We validated the approach on two publicly available databases, and we report up to 60% (Rimes) and 35% (IAM) relative improvement compared to a Gaussian HMM based on pixel values. The final systems give comparable to recurrent neural networks, which are the best systems since 2009.

Moysset, Bastien; Kermorvant, Christopher
On the evaluation of handwritten text line detection algorithms inproceedings
International Conference of Document Analysis and Recognition , 2013

Links (2)

Kermorvant, Hervé Glotin Xanadu Halkias Christopher
General Sparse Penalty in Deep Belief Networks: Towards Domain Adaptation techreport
2013

Bluche, Théodore; Ney, Hermann; Kermorvant, Christopher
Feature extraction with convolutional neural networks for handwritten word recognition inproceedings
International Conference of Document Analysis and Recognition , 2013

Links (2)

Menasri, Farès; Louradour, Jérôme; Bianne-Bernard, Anne-laure; Kermorvant, Christopher
The A2iA French handwriting recognition system at the Rimes-ICDAR2011 competition inproceedings
Document Recognition and Retrieval Conference , 2012

Bianne-bernard, Anne-laure; Menasri, Fares; Likforman-sulem, Laurence; Mokbel, Chafic; Kermorvant, Christopher
Variable length and context-dependent HMM letter form models for Arabic handwritten word recognition inproceedings
Document Recognition and Retrieval Conference , 2012

Louradour, Christopher Kermorvant Anne-Laure Bianne-Bernard Théodore Bluche Jérôme
On using alternative recognition candidates and scores for handwritten documents classification techreport
2012

Louradour, Jérôme; Bluche, Théodore; Bianne-Bernard, Anne-Laure; Menasri, Farès; Kermorvant, Christopher
De l'usage des scores et des alternatives de reconnaissance pour la classification d'images de documents manuscrits inproceedings
Colloque International Francophone sur l'Ecrit et le Document , 2012

Likforman-Sulem, Laurence; Hajj, Ramy Mohammad Al; Mokbel, Chafic; Menasri, Fares; Bianne-Bernard, Anne-Laure; Kermorvant, Christopher
Features for HMM-Based Arabic Handwritten Word Recognition Systems incollection
Märgner, Volker; El Abed, Haikal , Springer , Guide to OCR for Arabic Scripts , 2012

Bianne, A.-L.; Menasri, F.; Al-Hajj, R.; Mokbel, C.; Kermorvant, C.; Likforman-Sulem, L.
Dynamic and Contextual Information in HMM modeling for Handwriting Recognition article
IEEE Trans. on Pattern Analysis and Machine Intelligence , 2011

Sibade, Cédric; Retornaz, Thomas; Nion, Thibauld; Lerallut, Romain; Kermorvant, Christopher
Automatic indexing of French handwritten census registers for probate genealogy inproceedings
First International Workshop on Historical Document Imaging and Processing , 2011

Bianne-Bernard, A.-L.; Kermorvant, C.; Likforman-Sulem, L.; Mokbel, C.
Modélisation de HMMs en contexte avec des arbres de décision pour la reconnaissance de mots manuscrits article
Document Numérique , 2011

Louradour, Jérôme; Kermorvant, Christopher
Sample-dependent feature selection for faster document image categorization inproceedings
International Conference on Document Analysis and Recognition , 2011

Kermorvant-Duchemin, Elsa; Iacobelli, Silvia; Dit-Trolli, Sergio Eleni; Bonsante, Francesco; Kermorvant, Christopher; Sarfati, Gilles; Gouyon, Jean-Bernard; Lapillonne, Alexandre
Early chloride intake does not parallel that of sodium in extremely low birth weight infants and may impair neonatal outcomes. article
Journal of pediatric gastroenterology and nutrition , 2011

Abstract

BACKGROUND:: Accurate data on the optimal chloride intake in premature infants are scarce. OBJECTIVE:: To describe chloride (Cl) intakes in the first 10 days (D) of life and to assess the relationships between high Cl intakes and corrected serum chloride level or markers of severe acidosis in infants less than 28 weeks gestation. METHODS:: Retrospective cohort study including all infants < 28 weeks admitted to the neonatal intensive care unit over a 3-year period and cared for from birth until D10 or more. RESULTS:: Fifty-six infants were included. Cumulative total Cl intakes reached (mean +/- SD) 9.6 +/- 3.7 mmol/kg at day 3 and 49.2 +/- 13.5 mmol/kg at D10. Inadvertent intakes (from intravenous fluids other than parenteral nutrition) represented on average 70% of total Cl intakes in the first 3 days. Difference between Cl and sodium intakes reached (mean +/- SD) 7.8 +/- 4.8 mmol/kg at D10 and mainly originated from parenteral nutrition. By multivariate analysis, cumulative Cl intake > 10 mmol/kg during the first 3 days was an independent risk factor of base excess <-10 mmol/l. Cumulative Cl intake > 45 mmol/kg during the first 10 days was an independent risk factor of corrected chloremia > 115 mmol/l and of base excess <-10 mmol/l. CONCLUSIONS:: Cumulative Cl intake over 10 mmol/kg during the first 3 days (i.e. 3.3 mmol/kg/d on average) and over 45 mmol/kg during the first 10 days (i.e. 4.5 mmol/kg/d on average) may have unwanted metabolic consequences and should be avoided. Imbalance between electrolytes provided by the parenteral nutrition solution need to be detected and corrected.

Kermorvant, C.; Menasri, F.; Bianne, A.-L.; Al-Hajj, R.; Mokbel, C.; Likforman-Sulem, L.
The A2iA-Telecom ParisTech-UOB System for the ICDAR 2009 Handwriting Recognition Competition. inproceedings
Proc. of the Int. Conf. on Frontiers in Handwriting Recognition , 2010

Bianne, Anne-Laure; Kermorvant, Christopher; Likforman-sulem, Laurence; Mokbel, Chafic
Modélisation de HMMs en contexte avec des arbres de décision pour la reconnaissance de mots manuscrits inproceedings
Colloque International Francophone sur l'Ecrit et le Document , 2010

Kermorvant, C.; Louradour, J.
Handwritten mail classification experiments with the Rimes database inproceedings
Proc. of the Int. Conf. on Frontiers in Handwriting Recognition , 2010

Kermorvant, C.; Louradour, J.
How to save feature extraction time for fast and robust classification? inproceedings
Proceedings of the Int. Conf. on Artificial Intelligence and Statistics , 2010

Bianne, Anne-Laure; Kermorvant, Christopher; Likforman-Sulem, Laurence
Context-dependent HMM modeling using tree-based clustering for the recognition of handwritten words inproceedings
Document Recognition and Retrieval Conference , 2010

Bianne, Anne-Laure; Kermorvant, Christopher; Marty, Patrick; Menasri, Farès
Les caractères ne sont pas la clef des champs inproceedings
Conférence Francophone sur l'Apprentissage Automatique , 2009

Kermorvant, C.; Bianne, A.-L.; Marty, Patrick; Menasri, Farès
From Isolated Handwritten Characters to Fields Recognition: There's Many a Slip twixt Cup and Lip inproceedings
International Conference on Document Analysis and Recognition , 2009

Kermorvant, C.
Some Hints on the Teaching of Machine Learning to Industrial Practitioners inproceedings
Teaching Machine Learning: workshop on open problems and new directions , 2008

Kermorvant, Christopher; Bianne-Bernard, Anne-Laure; Bluche, Théodore; Louradour, Jérôme
On using alternative recognition candidates and scores for handwritten documents classi- fication techreport
2008

Kermorvant, C.; Rafrafi, A.
Automata Learning for Numerical Entities Extraction from OCR output inproceedings
Workshop on Challenges and Applications of Grammar Induction , 2007

Kermorvant, Christopher; Bengio, Yoshua
Extraction de sens à partir de bi-textes techreport
2004

Kermorvant, Christopher; Higuera, Colin; Dupont, Pierre
Learning typed automata from automatically labeled data article
Journal Electronique d'Intelligence Artificielle , 2004

Kermorvant, Christopher; Higuera, C De La
Improving probabilistic automata learning with additional knowledge inproceedings
International Workshops on. Structural and Syntactic Pattern Recognition , 2004

Kermorvant, Christopher; Dupont, Pierre
Chaînes de Markov d'ordre variable pour la détection de domaines dans les protéines inproceedings
Journées ouvertes biologie informatique mathématique , 2002

Kermorvant, Christopher; Higuera, Colin De
Learning Languages with Help inproceedings
International Colloquium on Grammatical Inference , 2002

Kermorvant, Christopher; Dupont, Pierre
Improved smoothing for probabilistic suffix trees seen as variable order Markov chains inproceedings
Springer , European Conference on Machine Learning , 2002

Higuera Pierre Dupont, Christopher Kermorvant Colin
Construction de Modèles de Langages par Inférence d’Automates Typés à partir de Données Etiquetées Automatiquement conference
Conférence Francophone sur l'Apprentissage Automatique , 2002

Kermorvant, Christopher; Dupont, Pierre
Stochastic grammatical inference with multinomial tests inproceedings
International Colloquium on Grammatical Inference , 2002

Kermorvant, Christopher; Dupont, Pierre
Inférence d’automates et correction d’erreur pour la classification des protéines inproceedings
Conférence Francophone sur l'Apprentissage Automatique , 2001

Kermorvant, C.; Morris, A.
A comparison of two strategies for ASR in additive noise : Missing Data and Spectral Subtraction techreport
Sixth European Conference on Speech Communication and Technology , 1999

Kermorvant, C.
A comparison of noise reduction techniques for robust speech recognition techreport
IDIAP , 1999

Abstract

This report presents the integration of several noise reduction methods into the front-end for speech recognition developed at IDIAP. The chosen methods are : Spectral Subtraction, Cepstral Mean Subtraction and Blind Equalization. These different methods are studied from a theoretical point of view, their implementation is described and are tested on the Numbers95 speech database. A good noise robustness is obtained by combining two of these methods, like Spectral Subtraction with Cepstral Mean Subtraction or Spectral Subtraction with Blind Equalization. The later combination is found to be more appropriate for real recognition systems since it is frame synchronous. A comparison with Jah-RASTA-PLP is also given.

Kermorvant, C.; Mokbel, C.
Towards introducing long-term statistics in MUSE for robust speech recognition inproceedings
IDIAP , Workshop on Automatic Speech Recognition and Understanding , 1999