Zero-Shot Learning Based Approach For Medieval Word Recognition using Deep-Learned Features

Chanda, Sukalpa; Baas, Jochem; Haitink, Daniel; Hamel, Sébastien; Stutzmann, Dominique; Schomaker, Lambert

doi:10.1109/icfhr-2018.2018.00067

Cited by 15 publications

(15 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The ensemble of five networks improves the results, as in [18]. For future work, we want to extend our approach to address the line recognition task and Zero-shot learning [34].…”

Section: B Quantitative Resultsmentioning

confidence: 85%

Improving the robustness of LSTMs for word classification using stressed word endings in dual-state word-beam search

Ameryan

Schomaker

2020

2020 17th International Conference on Frontiers in Handwriting Recognition (ICFHR)

Self Cite

View full text Add to dashboard Cite

In recent years, long short-term memory neural networks (LSTMs) followed by a connectionist temporal classification (CTC) have shown strength in solving handwritten text recognition problems. Such networks can handle not only sequence variability but also geometric variation by using a convolutional front end, at the input side. Although different approaches have been introduced for decoding activations in the CTC output layer, only limited consideration is given to the use of proper label-coding schemes. In this paper, we use a limited-size ensemble of end-to-end convolutional LSTM Neural Networks to evaluate four label-coding schemes. Additionally, we evaluate two CTC search techniques: Bestpath search vs dual-state word-beam search (DSWBS). The classifiers in the ensemble have comparable architectures but variable numbers of hidden units. We tested the coding and search approaches on three datasets: A standard benchmark IAM dataset (English) and two more difficult historical handwritten datasets (diaries and field notes, highly multilingual). Results show that stressing the word endings in the labelcoding scheme yields a higher performance, especially for DSWBS. However, stressing the start-of-word shapes with a token appears to be disadvantageous.

show abstract

“…The ensemble of five networks improves the results, as in [18]. For future work, we want to extend our approach to address the line recognition task and Zero-shot learning [34].…”

Section: B Quantitative Resultsmentioning

confidence: 85%

Improving the robustness of LSTMs for word classification using stressed word endings in dual-state word-beam search

Ameryan

Schomaker

2020

2020 17th International Conference on Frontiers in Handwriting Recognition (ICFHR)

Self Cite

View full text Add to dashboard Cite

show abstract

“…Word recognition in a one-shot learning framework and Zero-Shot learning framework has been investigated in [12,13] respectively. In [12], authors proposed a modified Siamese network to classify the text document in word level in a oneshot learning framework.…”

Section: Related Workmentioning

confidence: 99%

“…In [12], authors proposed a modified Siamese network to classify the text document in word level in a oneshot learning framework. In [13] the authors proposed a novel method that can classify completely unseen class images with good accuracy.…”

Section: Related Workmentioning

confidence: 99%

Bengali Place Name Recognition - Comparative Analysis Using Different CNN Architectures

Prasad¹,

Banerjee²,

Chanda

et al. 2021

Communications in Computer and Information Science

Self Cite

View full text Add to dashboard Cite

Optical Character Recognition (OCR) has been deployed in the past in different application areas such as automatic transcription and indexing of document images, reading aid for the visually impaired persons, postal automation etc. However, the performance in many cases has not been impressive due to the fact that character segmentation is itself an error-prone and difficult operation, which leads to the poor performance of the system due to erroneous segmentation of characters. Hence, for many applications (like document indexing, Postal automation) where full character-wise transcription is not required, word recognition is the preferred method these days. This article investigates recognition of Bengali place names as word images using 5 different traditional architectures. Experiments on word images (of Bengali place names) from 608 classes were conducted. Encouraging results were obtained in all instances. Keywords: Bengali word image recognition • Bengali postal address recognition • Bengali script • Bengali word image datasetThe data-set will be available on request. Please contact the first author via email.

show abstract

“…In [23], the proposed method uses a deep recurrent neural network (RNN) and a statistical character language model to attain high accuracy in terms of word spotting and word indexing. Only [1] explored very briefly Latin script word recognition problem in a ZSL framework. From the brief discussion, it is evident that though Zeroshot learning has been used extensively for animal, object and scenery image classification, it has been never used for word/text classification of any Indic script.…”

Section: Related Workmentioning

confidence: 99%

“…Encouraging results were obtained in a five-fold cross-validation setup. To the best of our knowledge only one article [1] explored word recognition in a Zero-Shot Learning framework that deals with Latin word images. But this is the very first work on any Indic script word recognition in a Zero-Shot Learning framework.…”

Section: Introductionmentioning

confidence: 99%

Recognizing Bengali Word Images - A Zero-Shot Learning Perspective

Chanda

Haitink

Prasad

et al. 2021

2020 25th International Conference on Pattern Recognition (ICPR)

Self Cite

View full text Add to dashboard Cite

Zero-Shot Learning(ZSL) techniques could classify a completely unseen class, which it has never seen before during training. Thus, making it more apt for any real-life classification problem, where it is not possible to train a system with annotated data for all possible class types. This work investigates recognition of word images written in Bengali Script in a ZSL framework. The proposed approach performs Zero-Shot word recognition by coupling deep learned features procured from various CNN architectures along with 13 basic shapes/stroke primitives commonly observed in Bengali script characters. As per the notion of ZSL framework those 13 basic shapes are termed as "Signature/Semantic Attributes". The obtained results are promising while evaluation was carried out in a Five-Fold cross-validation setup dealing with samples from 250 word classes.

show abstract

Zero-Shot Learning Based Approach For Medieval Word Recognition using Deep-Learned Features

Abstract: L. (2018). Zero-shot learning based approach for medieval word recognition using deep-learned features. In Proceedings of International Conference on Frontiers in Handwriting Recognition, ICFHR (pp. 345-350).

Cited by 15 publications

References 14 publications

Improving the robustness of LSTMs for word classification using stressed word endings in dual-state word-beam search

Improving the robustness of LSTMs for word classification using stressed word endings in dual-state word-beam search

Bengali Place Name Recognition - Comparative Analysis Using Different CNN Architectures

Recognizing Bengali Word Images - A Zero-Shot Learning Perspective

Contact Info

Product

Resources

About