Premium
Extraction of Character Sequences from Signboards by Using Background Color Information
Author(s) -
NOMURA MATSUNOBU,
SUZUKI TAKUMA,
KAGEYAMA YOICHI,
ISHIZAWA CHIKAKO,
NISHIDA MAKOTO
Publication year - 2015
Publication title -
electronics and communications in japan
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.131
H-Index - 13
eISSN - 1942-9541
pISSN - 1942-9533
DOI - 10.1002/ecj.11730
Subject(s) - character (mathematics) , computer science , artificial intelligence , computer vision , sequence (biology) , image processing , image (mathematics) , pattern recognition (psychology) , mathematics , genetics , geometry , biology
SUMMARY Smart phones, tablets, and mobile phones with built‐in cameras are very popular today because they are easy to use. It is important for image processors to be able to identify a signboard in an image and extract the character sequence. Techniques for extracting a character sequence from a signboard can be combined with various other systems to realize a diverse range of services. When utilization of such a system is assumed, it is necessary to take into consideration improvements in the accuracy of extraction of character sequence regions and improvements in processing speed. Therefore, this paper proposes an algorithm for extracting a character sequence from a signboard on the basis of scene image features. In order to improve the processing speed and accuracy, information about the background color of the signboard obtained in cooperation with a car navigation system or any other source was also utilized. In an experiment conducted on 319 color signboards, the proposed method detected 313 signboards (98.12%). Among 569 black and white signboards, the proposed method detected 513 signboards (90.16%). The average processing time in the proposed method was about 0.47 s for a color signboard and about 0.95 s for a black and white signboard. The experimental results revealed that the proposed method could accurately extract various types of character strings from signboards.