Natural Scene Text Detection Based on Multi-Channel FASText

DOI

10.2991/icacie-17.2017.4How to use a DOI?

Keywords

multi-channel; FASText; natural scene; text detection

Abstract

In view of the complexity of background and the variety of text in natural scene images, a multi-channel FASText based text detection method for natural scene images is proposed. To detect more texts as much as possible, the character candidates are extracted by proposed multi-channel FASText algorithm from the R, G and B component image respectively. Then, texture features of the character candidates are extracted to train a random forest character classifier and the non-characters are eliminated. At last, the character regions are merged into text regions according to the color distance feature and geometric adjacency feature. The proposed approach on ICDAR 2013 dataset achieves 76.76%, 80.17%, and 78.43% in recall rate, precision rate and f-score respectively. Compared to other state-of-the-art methods, both the recall rate and f-score are improved. Experimental result shows that the proposed method is effective to natural scene text detection.

Copyright

© 2017, the Authors. Published by Atlantis Press.

Open Access

This is an open access article distributed under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/).

Download article (PDF)

TY  - CONF
AU  - Chenfeng Guo
AU  - Juhua Liu
PY  - 2017/08
DA  - 2017/08
TI  - Natural Scene Text Detection Based on Multi-Channel FASText
BT  - Proceedings of the 2017 2nd International Conference on Automatic Control and Information Engineering (ICACIE 2017)
PB  - Atlantis Press
SP  - 16
EP  - 20
SN  - 2352-5401
UR  - https://doi.org/10.2991/icacie-17.2017.4
DO  - 10.2991/icacie-17.2017.4
ID  - Guo2017/08
ER  -