Real-Time Camera-Based Character Recognition

Japanese page is here (more informative)

An Application of Real-Time Camera-Based Character Recognition

Last update

11 May, 2011

Overview

This is a technique to recognize camera-captured words in a scene in real time. This is the first one which satisfies the following three requirements: ready for real-time processing, robust to perspective distortion, free from layout constraints.

Using the technique, we developed a prototype system of a new interface that works on a normal laptop PC (See the figure above). For a word pointed with a camera, the system shows the translation of the word, an image and a sound related to the word. For instance, when the word "Hawk" is recognized, it displays hawk's image or plays hawk's cry. With the interface, the user can acquire information easily such like opening a web page by clicking a link.

Movies

  1. A Camera-Based Interface Which Makes Texts in Scenes Anchor Texts

  2. WMV ( 35MB)

    light-weight WMV (7.6MB)

Presentations of the Method

  1. Real-Time Camera-Based Character Recognition Free from Layout Constraintsl (CBDAR2009)

  2. PPTX (3.4MB)

    PPT (6.0MB)

    PDF (2.4MB)

Other Information

Publications in English

SPIE Newsroom
Real-life clickable text
DAS2010
Memory-Based Recognition of Camera-Captured Characters
This paper won IAPR Nakano Award (Best Paper Award)
ICDAR2009
Real-Time Camera-Based Recognition of Characters and Pictograms
CBDAR2009
Real-Time Recognition of Camera-Captured Characters in Complex Layouts

Newspapers

THE NIKKAN KOGYO SHIMBUN
July 24, 2009 in Japanese

Patents

Name
Pattern Recognition Device
Applicant
Public University Cooperation Osaka Prefecture University
Inventors
Masakazu Iwamura and Koichi Kise
Application
2009-29031, JP (Feb. 10, 2009)
Priority Application
2009-163924, JP (July 10, 2009)
Int'l Application
PCT/JP2010/51889 (Feb. 9, 2010)
Publication
WO/2010/092952 (Aug. 19, 2010)

People

--> Homepage of Intelligent Media Processing Laboratory