Detail publikace

Handwriting Comenia Script Recognition with Convolutional Neural Network

Originální název

Handwriting Comenia Script Recognition with Convolutional Neural Network

Anglický název

Handwriting Comenia Script Recognition with Convolutional Neural Network

Jazyk

en

Originální abstrakt

This paper deals with handwriting recognition (HWR) using artificial intelligence of so–called Comenia script - a modern handwritten font similar to block letters recently introduced at primary schools in the Czech Republic. This work describes a method how to extend a limited training set of handwritten letters and proposes a new method to increase stability and accuracy by artificially created image samples. We examined a large set of algorithms including a deep learning method for classification of the handwriting characters. The best results were achieved using a convolutional neural network, which achieved the accuracy or character recognition 90.04%

Anglický abstrakt

This paper deals with handwriting recognition (HWR) using artificial intelligence of so–called Comenia script - a modern handwritten font similar to block letters recently introduced at primary schools in the Czech Republic. This work describes a method how to extend a limited training set of handwritten letters and proposes a new method to increase stability and accuracy by artificially created image samples. We examined a large set of algorithms including a deep learning method for classification of the handwriting characters. The best results were achieved using a convolutional neural network, which achieved the accuracy or character recognition 90.04%

BibTex


@inproceedings{BUT137770,
  author="Martin {Rajnoha} and Radim {Burget} and Malay Kishore {Dutta}",
  title="Handwriting Comenia Script Recognition with Convolutional Neural Network",
  annote="This paper deals with handwriting recognition (HWR) using artificial intelligence of so–called Comenia script - a modern handwritten font similar to block letters recently introduced at primary schools in the Czech Republic. This work describes a method how to extend a limited training set of handwritten letters and proposes a new  method to increase stability and accuracy by artificially created image samples. We examined a large set of algorithms including a deep learning method for classification of the handwriting characters. The best results were achieved using a convolutional neural network, which achieved the accuracy or character recognition 90.04%",
  booktitle="40th Anniversary of International Conference on Telecommunications and Signal Processing (TSP)",
  chapter="137770",
  doi="10.1109/TSP.2017.8076093",
  howpublished="electronic, physical medium",
  year="2017",
  month="july",
  pages="775--779",
  type="conference paper"
}