A set of benchmarks for Handwritten Text Recognition on historical documents

Sánchez, Joan-Andreu; Romero, Verónica; Toselli, Alejandro H.; Villegas, Mauricio; Vidal, Enrique

doi:10.1016/j.patcog.2019.05.025

Cited by 70 publications

(50 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…erefore, many experts and scholars have studied the intelligent and efficient methods of English feature recognition [6]. According to the differences in text information, scholars put forward targeted improvement strategies for the recognition of different text information [7].…”

Section: Introductionmentioning

confidence: 99%

[Retracted] Feature Recognition of English Based on Deep Belief Neural Network and Big Data Analysis

Liu

2021

Computational Intelligence and Neuroscience

View full text Add to dashboard Cite

Realizing accurate recognition of Chinese and English information is a major difficulty in English feature recognition. Based on this difficulty, this paper studies the English feature recognition model based on deep belief network classification algorithm and Big Data analysis. First, the basic framework based on deep belief network classification algorithm and Big Data analysis is proposed. Combined with the Big Data analysis training model, the English feature information is processed. Through the recognition of different English text features, the recognition and matching of English features are realized. Then the errors of deep belief network classification algorithm and Big Data analysis are evaluated. Second, this paper describes the quantitative evaluation of deep belief network classification algorithm and Big Data analysis in this system. In the evaluation, the language feature evaluation method is used to improve the evaluation function. At the same time, the deep belief network classification algorithm and Big Data analysis are used to self-study the model, and the English feature recognition method with strong applicability is established. Finally, the effectiveness of the recognition system is verified by the experiment.

show abstract

Section: Introductionmentioning

confidence: 99%

[Retracted] Feature Recognition of English Based on Deep Belief Neural Network and Big Data Analysis

Liu

2021

Computational Intelligence and Neuroscience

View full text Add to dashboard Cite

show abstract

“…Actually, to the best of our knowledge, no existing method works in an end-to-end fashion for full-page images. This holds for others fields with structurally simpler documents as well, such as optical recognition from text documents [33]. On the other hand, due to how the neural networks operate, the output is restricted to sequences, so the approach is limited to monophonic or homophonic music, as demonstrated in [2].…”

Section: Methodsmentioning

confidence: 99%

Evaluating Simultaneous Recognition and Encoding for Optical Music Recognition

Ríos-Vila

Calvo-Zaragoza

Rizo

2020

7th International Conference on Digital Libraries for Musicology

View full text Add to dashboard Cite

Most Optical Music Recognition workflows include several steps to retrieve the content from music score images. These steps typically comprise preprocessing, recognition, notation reconstruction and encoding. Currently, state-of-the-art models allow performing graphic recognition in an almost end-to-end fashion, performing the steps from preprocessing to recognition simultaneously. However, this graphic recognition has to be further processed to obtain a standard digital music representation. In this paper, we study the simultaneous recognition and encoding for a state-of-the-art OMR approach, based on neural networks, which receives a single staff-region image as input and directly obtains a sequence of characters that encodes the content in a standard music format. Our results confirm that performing OMR this way is feasible and brings additional benefits such as directly obtaining a version of the score readily available to be further processed or edited by standard tools. CCS CONCEPTS • Computing methodologies → Neural networks.

show abstract

“…The Handwritten Text Recognition (HTR) research field has the challenge of offering robust systems to deal with different scenarios in production environments, which usually have the handwritten sequence surrounded by background noise, or even combined with other irrelevant information [2], [40]. Therefore, state-of-the-art optical models in HTR fit with the objective of the HDSR research field presented above.…”

Section: State-of-the-art In Htrmentioning

confidence: 99%

HDSR-Flor: A Robust End-to-End System to Solve the Handwritten Digit String Recognition Problem in Real Complex Scenarios

et al. 2020

View full text Add to dashboard Cite

Automatic handwriting recognition systems are of interest for academic research fields and for commercial applications. Recent advances in deep learning techniques have shown dramatic improvement in relation to classic computer vision problems, especially in Handwritten Text Recognition (HTR). However, several approaches try to solve the problem of deep learning applied to Handwritten Digit String Recognition (HDSR), where it has to deal with the low number of trainable data, while learning to ignore any writing symbol around the digits (noise). In this context, we present a new optical model architecture (Gated-CNN-BGRU), based on HTR workflow, applied to HDSR. The International Conference on Frontiers of Handwriting Recognition (ICFHR) 2014 competition on HDSR were used as baselines to evaluate the effectiveness of our proposal, whose metrics, datasets and recognition methods were adopted for fair comparison. Furthermore, we also use a private dataset (Brazilian Bank Check-Courtesy Amount Recognition), and 11 different approaches from the state-of-the-art in HDSR, as well as 2 optical models from the state-of-the-art in HTR. Finally, the proposed optical model demonstrated robustness even with low data volume (126 trainable data, for example), surpassing the results of existing methods with an average precision of 96.50%, which is equivalent to an average percentage of improvement of 3.74 points compared to the state-of-the-art in HDSR. In addition, the result stands out in the competition's CVL HDS set, where the proposed optical model achieved a precision of 93.54%, while the best result so far had been from Beijing group (from the competition itself), with 85.29%.

show abstract

A set of benchmarks for Handwritten Text Recognition on historical documents

Cited by 70 publications

References 28 publications

[Retracted] Feature Recognition of English Based on Deep Belief Neural Network and Big Data Analysis

[Retracted] Feature Recognition of English Based on Deep Belief Neural Network and Big Data Analysis

Evaluating Simultaneous Recognition and Encoding for Optical Music Recognition

HDSR-Flor: A Robust End-to-End System to Solve the Handwritten Digit String Recognition Problem in Real Complex Scenarios

Contact Info

Product

Resources

About