Current state-of-the-art approaches to offline Handwritten Text Recognition extensively rely on Multidimensional Long Short-Term Memory networks However, these architectures come with quite an expensive computational cost, and we observe that they extract features visually similar to those of convolutional layers, which are computationally cheaper This suggests that the two-dimensional long-term dependencies, which are potentially modeled by multidimensional recurrent layers, may not be essential to achieve a good recognition accuracy, at least in the lower layers of the architecture In this work, an alternative model is explored that relies only on convolutional and one-dimensional recurrent layers that achieves better or equivalent results than those of the current state-of-the-art architecture, and runs significantly faster In addition, we observe that using random distortions during training as synthetic data augmentation dramatically improves the accuracy of our model Thus, are multidimensional recurrent layers really necessary for Handwritten Text Recognition? Probably not

Are Multidimensional Recurrent Layers Really Necessary for Handwritten Text Recognition

Multidimensional long short-term memory recurrent neural networks achieve impressive results for handwriting recognition. However, with current CPU-based implementations, their training is very expensive and thus their capacity has so far been limited. We release an efficient GPU-based implementation which greatly reduces training times by processing the input in a diagonal-wise fashion. We use this implementation to explore deeper and wider architectures than previously used for handwriting recognition and show that especially the depth plays an important role. We outperform state of the art results on two databases with a deep multidimensional network.

Handwriting Recognition with Large Multidimensional Long Short-Term Memory Recurrent Neural Networks

Given an image of a handwritten word, a CNN is employed to estimate its n-gram frequency profile, which is the set of n-grams contained in the word. Frequencies for unigrams, bigrams and trigrams are estimated for the entire word and for parts of it. Canonical Correlation Analysis is then used to match the estimated profile to the true profiles of all words in a large dictionary. The CNN that is used employs several novelties such as the use of multiple fully connected branches. Applied to all commonly used handwriting recognition benchmarks, our method outperforms, by a very large margin, all existing methods.

/pdf/cnn-n-gram-for-handwritingword-recognition-3vyy6rqsl2.pdf

CNN-N-Gram for HandwritingWord Recognition

In this paper we explore a new model focused on integrating two classifiers; Convolutional Neural Network (CNN) and Support Vector Machine (SVM) for offline Arabic handwriting recognition (OAHR) on which the dropout technique was applied. The suggested system altered the trainable classifier of the CNN by the SVM classifier. A convolutional network is beneficial for extracting features information and SVM functions as a recognizer. It was found that this model both automatically extracts features from the raw images and performs classification. Additionally, we protected our model against over-fitting due to the powerful performance of dropout. In this work, the recognition on the handwritten Arabic characters was evaluated; the training and test sets were taken from the HACDB and IFN/ENIT databases. Simulation results proved that the new design based-SVM of the CNN classifier architecture with dropout performs significantly more efficiently than CNN based-SVM model without dropout and the standard CNN classifier. The performance of our model is compared with character recognition accuracies gained from state-of-the-art Arabic Optical Character Recognition, producing favorable results.

/pdf/a-new-design-based-svm-of-the-cnn-classifier-architecture-6uq9iimqlj.pdf

A New Design Based-SVM of the CNN Classifier Architecture with Dropout for Offline Arabic Handwritten Recognition

We introduce two data augmentation and normalization techniques, which, used with a CNN-LSTM, significantly reduce Word Error Rate (WER) and Character Error Rate (CER) beyond best-reported results on handwriting recognition tasks (1) We apply a novel profile normalization technique to both word and line images (2) We augment existing text images using random perturbations on a regular grid We apply our normalization and augmentation to both training and test images Our approach achieves low WER and CER over hundreds of authors, multiple languages and a variety of collections written centuries apart Image augmentation in this manner achieves state-of-the-art recognition accuracy on several popular handwritten word benchmarks

Data Augmentation for Recognition of Handwritten Words and Lines Using a CNN-LSTM Network

The paper presents the French RIMES (Reconnaissance et Indexation de donnees Manuscrites et de fac similES / Recognition and Indexing of handwritten documents and faxes) project, which is an evaluation campaign dedicated to automatic systems of recognition and indexing of handwritten letters such as those sent by postal mail or fax by individuals to companies or administrations. This project is intended to all the key players of the handwritten recognition and document analysis fields by proposing them to participate to evaluation tests, a first test having taken already place. In this framework, a new database of more than 5600 real mails completely annotated have been created as well as secondary databases of isolated characters, handwritten words (300 000 snippets) and logos. Beyond the evaluation, the other contribution of the RIMES project is to make available all these databases to the community after the last test

Emmanuel Augustin

Papers

RIMES evaluation campaign for handwritten mail processing