A Novel Connectionist System for Unconstrained Handwriting Recognition

Abstract

Abstract-Recognising lines of unconstrained handwritten text is a challenging task. The difficulty of segmenting cursive or overlapping characters, combined with the need to exploit surrounding context, has led to low recognition rates for even the best current recognisers. Most recent progress in the field has been made either through improved preprocessing, or through advances in language modelling. Relatively little work has been done on the basic recognition algorithms. Indeed, most systems rely on the same hidden Markov models that have been used for decades in speech and handwriting recognition, despite their well-known shortcomings. This paper proposes an alternative approach based on a novel type of recurrent neural network, specifically designed for sequence labelling tasks where the data is hard to segment and contains long range, bidirectional interdependencies. In experiments on two large unconstrained handwriting databases, our approach achieves word recognition accuracies of 79.7% on online data and 74.1% on offline data, significantly outperforming a state-of-the-art HMM-based system. In addition, we demonstrate the network's robustness to lexicon size, measure the individual influence of its hidden layers, and analyse its use of context. Lastly we provide an in depth discussion of the differences between the network and HMMs, suggesting reasons for the network's superior performance.

ICFHR2014 Competition on Handwritten Text Recognition on Transcriptorium Datasets (HTRtS)

“…Therefore, holistic, segmentation-free Handwritting Text Recognition (HTR) techniques are needed which not require any explicit character or word segmentation. Current technology for HTR borrows concepts and methods from the field of Automatic Speech Recognition, such as Hidden Markov 1 http://www.transcriptorium.eu/ Models (HMMs), Neural Networks and N-grams [2], [3], [4]. These models are trained from samples by using efficient techniques.…”

Confidence- and margin-based MMI/MPE discriminative training for off-line handwriting recognition

“…Interesting for further research will remain hybrid HMM/ANN approaches [13,10], combining the advantages of large and non-linear context modeling via neural networks while profiting from the Markovian sequence modeling. This is also supported by the 36% relative improvement we could achieve in the ICFHR 2010 Arabic handwriting competition [26] with the proposed framework but an MLP based feature extraction.…”

Character-Based Handwritten Text Recognition of Multilingual Documents

“…Our hypothesis is that tasks tackled in previous work did not contain a significant number of OOV words compared to the figures of the GERMANA database 1 . In GERMANA, the problem of OOV words is aggravated by its multilingual nature, since the presence of languages such as Latin, French, German and Italian is less than 4% of the total number of words.…”

“…To alleviate this effort, automatic handwriting transcription techniques based on speech recognition technology have flourished over the last years, although the quality of the transcriptions provided by these techniques is still far from not being in need of supervision [1]. An effective approach to supervision is to integrate an ongoing retraining system that interactively incorporates user corrections once a line has been reviewed.…”