Thesis Details

Neuronové sítě typu Transformer pro přepis ručně psaného textu

Master's Thesis Student: Vešelíny Peter Academic Year: 2021/2022 Supervisor: Kohút Jan, Ing.
Language
Slovak
Abstract

This Master's thesis aims to design a system using the transformer neural network and perform experiments with this proposed model in the task of handwriting text recognition. In this thesis, a multilingual dataset with predominate Czech texts is used. The experiments examine the influence of basic hyperparameters, such as network size, convolutional encoder type, and the use of different text tokenizers. In this work, I also use text corpora of the Czech language which is used to train the network decoder. Furthermore, I experiment with the usage of additional textual information during the decoding process. This information comes from the previous line of the transcribed image. The transformer achieves a character recognition error rate of 3.41 % on the test data set which is 0.16 % worse performance than the recurrent neural network achieves. To compare this model with other transformer-based models from available articles, the network was trained on the IAM dataset, where it achieved an error of 2.48 % and therefore outperformed other models in handwriting text recognition task.

Keywords

text recognition, handwriting text, neural networks, attention, transformer, text corpus

Department
Degree Programme
Files
Status
defended, grade B
Date
21 June 2022
Reviewer
Committee
Černocký Jan, prof. Dr. Ing. (DCGM FIT BUT), předseda
Burget Lukáš, doc. Ing., Ph.D. (DCGM FIT BUT), člen
Češka Milan, doc. RNDr., Ph.D. (DITS FIT BUT), člen
Hradiš Michal, Ing., Ph.D. (DCGM FIT BUT), člen
Rozman Jaroslav, Ing., Ph.D. (DITS FIT BUT), člen
Zbořil František V., doc. Ing., CSc. (DITS FIT BUT), člen
Citation
VEŠELÍNY, Peter. Neuronové sítě typu Transformer pro přepis ručně psaného textu. Brno, 2022. Master's Thesis. Brno University of Technology, Faculty of Information Technology. 2022-06-21. Supervised by Kohút Jan. Available from: https://www.fit.vut.cz/study/thesis/24792/
BibTeX
@mastersthesis{FITMT24792,
    author = "Peter Ve\v{s}el\'{i}ny",
    type = "Master's thesis",
    title = "Neuronov\'{e} s\'{i}t\v{e} typu Transformer pro p\v{r}epis ru\v{c}n\v{e} psan\'{e}ho textu",
    school = "Brno University of Technology, Faculty of Information Technology",
    year = 2022,
    location = "Brno, CZ",
    language = "slovak",
    url = "https://www.fit.vut.cz/study/thesis/24792/"
}
Back to top