BLSTM-CTC Combination Strategies for Off-line Handwriting Recognition

Luc Mioulet (1) , G. Bideault (2) , C. Chatelain (1) , T. Paquet (2) , S. Brunessaux (2)

1 UNIROUEN - Université de Rouen Normandie
2 Airbus [France]

In this paper we present several combination strategies using multiple BLSTM-CTC systems. Given several feature sets our aim is to determine which strategies are the most relevant to improve on an isolated word recognition task (the WR2 task of the ICDAR 2009 competition), using a BLSTM-CTC architecture. We explore different combination levels: early integration (feature combination), mid level combination and late fusion (output combinations). Our results show that several combinations outperform single feature BLSTM-CTCs.

Keywords: Handwriting Recognition; Feature Combination; Recurrent Neural Network; Neural Network

Domains: Computer Science [cs]; Computer Vision and Pattern Recognition [cs.CV]

Luc Mioulet, G. Bideault, C. Chatelain, T. Paquet, S. Brunessaux. BLSTM-CTC Combination Strategies for Off-line Handwriting Recognition. International Conference on Pattern Recognition Applications and Methods, Jan 2015, Lisbon, France. pp.173-180, ⟨10.5220/0005178601730180⟩. ⟨hal-02087427⟩