| Autor: |
Fomin, S. A., Stupnikov, S. A. |
| Zdroj: |
Pattern Recognition & Image Analysis; Jun2023, Vol. 33 Issue 2, p92-100, 9p |
| Abstrakt: |
The recognition of a handwritten text is a classical computer science problem for which neural networks have recently been successfully applied. However, to achieve high accuracy, a modern neural network requires not only large computational resources but also a huge amount of data for learning. One of the most well-known ways to increase the amount of data for learning is augmentation, which is the generation of additional data from existing data. There are different ways to augment graphical data, varying from the simplest techniques such as image shifting to the most sophisticated methods in which data are generated using generative adversarial neural networks. This paper discusses various approaches to generating examples of human handwriting to improve the quality of handwriting recognition. Handwriting generation is a quite complex procedure that includes steps of generating characters, of combining them into words, and of final processing. There are many papers describing these stages for foreign languages, but not for Russian. The purpose of this paper is to summarize the current approaches in the field of handwriting generation and to develop an approach to the augmentation of human handwriting in the Russian language. [ABSTRACT FROM AUTHOR] |
| Databáze: |
Complementary Index |
| Externí odkaz: |
|
Full text from SpringerLink