Method and apparatus for recognition of character string in scene image

• US 7,403,656 B2
• Filed: 02/04/2005
• Issued: 07/22/2008
• Est. Priority Date: 08/20/2004
• Status: Expired due to Fees

• Current Assignee: Hitachi, Ltd.
• Original Assignee: Hitachi, Ltd.
• Inventors: Koga, Masashi
• Primary Examiner(s): Carter; Aaron W
• Application Number:
  US11/048,650
• Publication Number:
  US 20060039605A1
• Time in Patent Office: 1,264 Days
• Field of Search: 382173-216, 382229-231
• US Class Current: 382/177
• CPC Class Codes: G06V 20/63 Scene text, e.g. street names
• CPC Class Codes: G06V 30/10 Character recognition
• CPC Class Codes: G06V 30/162 Quantising the image signal

Abstract

A character recognition method that is robust under an unknown illumination condition is provided. An apparatus for realizing such robust character recognition includes plural different binarization, means for synthesizing character sub-image candidates that have been obtained from the binarization units, and means for analyzing character sub-image candidates and for recognizing an image as a character string consisting of character sub-image candidates.

Claims

1. An apparatus for recognition of a character string in a scene image, comprising:

a plurality of different binarization means for obtaining a plurality of different binary images from an input image;

pre-segmentation means for detecting a set of pixels as a set of character subimage candidates from each binary image obtained;

network synthesizing means for forming a single segmentation hypothesis network from a set of character subimage candidates obtained by the pre-segmentation means;

character string recognizing means that selects a set of character subimage candidates that is most plausible as a character from among a set of character subimage candidates of the segmentation hypothesis network, links a set of character subimage candidates on the hypothesis network, and extracts it as a character string;

text plane detecting means for detecting information of edges of a text plane from the input image; and

candidate character string selecting means for extracting a character string satisfying predetermined conditions using information of the edges of the text plane when a plurality of character string candidates are extracted as the character string.