The A2iA French handwriting recognition system at the Rimes-ICDAR2011 competition

Menasri, F.; Louradour, Jérôme; Bianne-Bernard, Anne-Laure; Kermorvant, Christopher

doi:10.1117/12.911981

Cited by 40 publications

(40 citation statements)

References 5 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, it is well known that combining systems or features may improve the overall results. 15,16 In this paper we explore different ways of combining an ensemble of F BLSTM-CTC. We explore different levels of information combination, from a low level combination (feature space combination), to mid-level combinations (internal system representation combinations), and high level combinations (decoding combinations).…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Exploring multiple feature combination strategies with a recurrent neural network architecture for off-line handwriting recognition

et al. 2015

View full text Add to dashboard Cite

The BLSTM-CTC is a novel recurrent neural network architecture that has outperformed previous state of the art algorithms in tasks such as speech recognition or handwriting recognition. It has the ability to process long term dependencies in temporal signals in order to label unsegmented data. This paper describes different ways of combining features using a BLSTM-CTC architecture. Not only do we explore the low level combination (feature space combination) but we also explore high level combination (decoding combination) and mid-level (internal system representation combination). The results are compared on the RIMES word database. Our results show that the low level combination works best, thanks to the powerful data modeling of the LSTM neurons.

show abstract

Section: Introductionmentioning

confidence: 99%

“…The composed system, referred as the BLSTM-CTC, has shown very impressive results on challenging databases. [12][13][14][15] In this paper, we present two baseline systems using the same architecture, built around the BLSTM-CTC network. These systems only differ by the features that are extracted using a sliding window method.…”

Section: Introductionmentioning

confidence: 99%

Exploring multiple feature combination strategies with a recurrent neural network architecture for off-line handwriting recognition

et al. 2015

View full text Add to dashboard Cite

show abstract

“…Since the foremost research was done in mid-1990s [1] [2] [5], the hidden Markov model (HMM) has become the state of the art offline handwriting recognition technique for alphabetic scripts such as Arabic, English and French [3] [4] [8] [6] [9] and French [6] [10]. The HMM provides a stochastic model for unified character segmentation and recognition.…”

Section: Introductionmentioning

confidence: 99%

Progress in the Raytheon BBN Arabic Offline Handwriting Recognition System

Cao

Natarajan

Peng

et al. 2014

2014 14th International Conference on Frontiers in Handwriting Recognition

View full text Add to dashboard Cite

This paper presents the most recent progress and state of the art result obtained from BBN's Arabic offline handwriting recognition research. Our system is based a left-toright hidden Markov model and integrates discriminative learning methods including discriminative MPE and n-best rescoring using the scores of glyph classifiers (SVM, DNN) and the RNNLM. Arabic-related features for n-best rescoring are also investigated in this paper. Multi-stage MAP/MLLR and writer verification are applied to adapt the recognizer in all training situations. Consensus network is extensively researched for system combination and improving challenging preprocessing problems. 123

show abstract

“…Using the RIMES data-set [16] (7400 words) and the letters extracted by A2iA [17] (first proposal) we test the influence of the size of the bigram set over the word recognition. It is important to note that the 'poor' quality of the letter extraction only allows a Word Recognition Rate of 28%.…”

Section: How Many Bigrams Per Word?mentioning

confidence: 99%

“…Since 2009, connectionist models such as multi-dimensional LSTM (Long Short-Term Memory) recurrent neural networks [9][10], deep feed-forward neural 1 , Christopher Kermorvant 2 , and Hervé Glotin 3 networks [11] and various mixtures of these have won several international connected handwriting competitions (such as the International Conference on Document Analysis and Recognition) without any prior knowledge about the various languages (French [17], Arabic [24]) to be learned. GPU-based deep learning methods for feed-forward networks were the first artificial pattern recognizers to achieve human-competitive performance [12] on the famous MNIST handwritten digits problem [13].…”

Section: Introductionmentioning

confidence: 99%

A Biologically Plausible SOM Representation of the Orthographic Form of 50,000 French Words

Touzet

Kermorvant²,

Glotin

2014

Advances in Self-Organizing Maps and Learning Vector Quantization

Self Cite

View full text Add to dashboard Cite

Abstract. Recently, an important aspect of human visual word recognition has been characterized. The letter position is encoded in our brain using an explicit representation of order based on letter pairs: the openbigram coding [15]. We hypothesize that spelling has evolved in order to minimize reading errors. Therefore, word recognition using bigramsinstead of letters -should be more efficient. First, we study the influence of the size of the neighborhood, which defines the number of bigrams per word, on the performance of the matching between bigrams and word. Our tests are conducted against one of the best recognition solutions used today by the industry, which matches letters to words. Secondly, we build a cortical map representation of the words in the bigram space -which implies numerous experiments in order to achieve a satisfactory projection. Third, we develop an ultra-fast version of the self-organizing map in order to achieve learning in minutes instead of months.

show abstract

The A2iA French handwriting recognition system at the Rimes-ICDAR2011 competition

Cited by 40 publications

References 5 publications

Exploring multiple feature combination strategies with a recurrent neural network architecture for off-line handwriting recognition

Exploring multiple feature combination strategies with a recurrent neural network architecture for off-line handwriting recognition

Progress in the Raytheon BBN Arabic Offline Handwriting Recognition System

A Biologically Plausible SOM Representation of the Orthographic Form of 50,000 French Words

Contact Info

Product

Resources

About