CNN based recognition of handwritten multilingual city names
Article Type
Research Article
Publication Title
Multimedia Tools and Applications
Abstract
It is important to recognize the destination city name correctly for a postal document to reach its desired address. In India people often mix up scripts while writing the address. Often the script of the destination city name is different from the other part of the postal document. This is common in India due to the multilingual and multi script nature of the country. In this paper, a Convolutional Neural Network (CNN) based approach towards the recognition of handwritten multilingual multiscript Indian city names is presented. Experiments were performed not only in a single script scenario but also in multi script, considering English, Bangla and Devanagari scripts. An accuracy of 91.72% was obtained on 106 city names in mixed script scenario from the proposed scheme and the data set will be made available to the researcher on request. Further experiments were also performed with different script combinations and obtained results up to 98.01%. The system also produced a mean performance difference of approximately ± 1% for successive changes in the data set size, thereby pointing to the robustness of the proposed architecture.
First Page
11501
Last Page
11517
DOI
10.1007/s11042-022-12193-8
Publication Date
3-1-2022
Recommended Citation
Roy, Ramit Kumar; Mukherjee, Himadri; Roy, Kaushik; and Pal, Umapada, "CNN based recognition of handwritten multilingual city names" (2022). Journal Articles. 3222.
https://digitalcommons.isical.ac.in/journal-articles/3222