Using the Web to Create Dynamic Dictionaries in Handwritten Out-of-Vocabulary Word Recognition

Abstract

Abstract-Handwriting recognition systems rely on predefined dictionaries obtained from training data. Small and static dictionaries are usually exploited to obtain high in-vocabulary (IV) accuracy at the expense of coverage. Thus the recognition of out-of-vocabulary (OOV) words cannot be handled efficiently. To improve OOV recognition while keeping IV dictionaries small, we introduce a multi-step approach that exploits Web resources. After an initial IV-OOV sequence classification, external resources are used to create OOV sequence-adapted dynamic dictionaries. A final Viterbi-based decoding is performed over the dynamic dictionary to determine the most probable word for the OOV sequence. We validate our approach with experiments conducted on RIMES, a publicly available database. Results show that improvements are obtained compared to standard handwriting recognition, performed with a static dictionary. Both domainadapted and generic dynamic dictionaries are studied and we show that domain adaptation is beneficial.

Comparative Study of HMM and BLSTM Segmentation-Free Approaches for the Recognition of Handwritten Text-Lines

“…It reaches a 16.1% WER. Both recognizers are challenged by the out-of-vocabulary words (OOVs) [20]. In this example, there is one OOV code sequence.…”

Study of the influence of lexicon and language restrictions on computer assisted transcription of historical manuscripts

“…On the other hand, HTR, as a natural language technology, has some limitations, such as those imposed by vocabulary constraints. One possible solution to overcome this limitation is the use of external textual resources to improve the linguistic models of the HTR system [20]. However, given the accuracy of HTR systems based on deep learning, it is usual to perform the recognition without lexical restrictions [21,22].…”

Reconnaissance de mots manuscrits hors-vocabulaire en utilisant des ressources webs

“…Les mots classés comme HV (leur score de log-probabilité normalisé Lp N < thre ) sont traités avec la méthode décrite dans la section 4. Dans (Oprean et al, 2013), nous avions montré que thre=mean HV maximise le taux de reconnaissance total. Lorsque thre=mean DV , 3 612 images de mots ont été classées comme HV, comparativement à 568 lorsque thre=mean HV .…”