System and method for identifying and extracting character strings from captured image data

• US 20030198386A1
• Filed: 04/19/2002
• Published: 10/23/2003
• Est. Priority Date: 04/19/2002
• Status: Abandoned Application

• Current Assignee: Hewlett-Packard Development Company, L.P. (HP Inc.)
• Original Assignee: Hewlett-Packard Development Company, L.P. (HP Inc.)
• Inventors: Luo, Huitao
• Application Number:
  US10/126,151
• Publication Number:
  US 20030198386A1
• Time in Patent Office: Days
• US Class Current: 382/199
• CPC Class Codes: G06V 20/62 Text, e.g. of license plates, overlay texts or captions on TV images
• CPC Class Codes: G06V 30/10 Character recognition

Abstract

A system and method of detecting and extracting a character string within captured image data of a natural image is described. An edge representation of the image data is obtained including a plurality of single width edge pixels each having an associated magnitude and direction. Edge pixels are labeled dependent pixel connectedness. End point edge pixels are identified and their labels are merged at high character curvature areas. Character bounding area definitions are created using the edge pixel direction information and labels. The character boundary definitions are filtered using direction information to identify character foreground and background information. Filtered definitions are merged with adjacent boundary definitions to form a character line definition dependent on the adjacent bounding area definitions relative location to each other.

Claims

1. A method of processing captured image data including image data corresponding to at least one string of characters comprising:

detecting edges within the image data to generate an edge representation having an associated plurality of edge pixels of a single pixel width, the edge representation including at least magnitude and direction information for each edge pixel;

assigning edge pixel labels to each edge pixel dependent on pixel connectedness;

creating bounding area definitions each corresponding to a character using direction information and dependent on similar edge pixel labels;

filtering bounding area definitions using direction information to identify character foreground and background information; and

merging filtered adjacent bounding area definitions to form a line definition dependent on the adjacent bounding area definitions relative location to each other.

2. The method as described in claim 1 wherein labeling edge pixels is dependent on identifying end point edge pixels.

3. The method as described in claim 1 wherein filtering using direction information further comprises scanning bounding area definitions with a projection line to identify pairs of edge pixels and associated gradient vectors of each pair of edge pixels to evaluate associated foreground and background information of the pair.

4. The method as described in claim 3 wherein filter using direction information further comprises statistically analyzing the gradient vector information for the pairs of edge pixels to identify a foreground shade associated with each bounding area definition.

5. The method as described in claim 1 wherein filtering further comprises analyzing geometric characteristics of the bounding area definitions.

6. The method as described in claim 1 wherein filtering further comprises identifying edge pixel pairs from direction information to identify the character background and foreground information.

7. The method as described in claim 1 wherein filtering further comprises analyzing grayscale characteristics of the bounding area definitions to identify the character background and foreground information.

8. The method as described in claim 1 wherein combining adjacent bounding area definitions to form a line definition comprises:

a) assigning a current bounding area definition to a current line definition;

b) searching a pre-defined area in the proximity of the current line;

c) locating a new bounding area at least overlapping the pre-defined area;

d) merging the definition of the new bounding area into the current line definition;

e) repeating b)-d) until no more new bounding areas.

9. The method as described in claim 1 further comprising binarizing each bounding area definition associated with each line definition by binarizing whole foreground and background segments of pixels for each bounding area definition.

10. A system of processing captured image data including image data corresponding to at least one string of characters comprising:

edge detector for detecting edges within the image data to generate an edge representation having an associated plurality of edge pixels of a single pixel width, the edge representation including at least magnitude and direction information for each edge pixel;

bounding area definition creator including at least an edge pixel labeler for assigning edge pixel labels to each edge pixel dependent on pixel connectedness and direction information;

bounding area definition filter for filtering bounding area definitions using direction information to identify character foreground and background information; and

line definition creator for merging filtered adjacent bounding area definitions to form a line definition dependent on the adjacent bounding area definitions relative location to each other.

11. The system as described in claim 10 wherein the bounding area definition creator further comprising:

end point detector, for identifying end points of high curvature character segments in the edge pixels; and

end point linker for linking corresponding identified end points by assigning the same labels.

12. The system as described in claim 10 wherein the bounding area filter further comprising:

topological analyzer and grayscale analyzer including means for scanning bounding area definitions with a projection line to identify pairs of edge pixels and associated gradient vectors of each pair of edge pixels to evaluate associated foreground and background information of the pair and dependent on the foreground and background information identifying non-erroneous bounding area definitions, wherein the grayscale analyzer using direction information to statistically analyze the gradient vector information for the pairs of edge pixels to identify a foreground color associated with each bounding area definition.

13. The system as described in claim 10 wherein filtering further comprises a geometric analyzer for filtering bounding area definitions by evaluating geometric characteristics of the bounding area definitions

14. The system as described in claim 10 wherein line definition creator comprises a positively identified bounding area definition merger and an “

other character”

element detector wherein merged positive definitions correspond to a given line definition and “

other character”

elements detected in a predefined area about the given line definition being merged into the line definition.

15. A computer readable medium containing a computer program for processing images wherein the computer program performs:

detecting edges within the image data to generate an edge representation having an associated plurality of edge pixels of a single pixel width, the edge representation including at least magnitude and direction information for each edge pixel;

assigning edge pixel labels to each edge pixel dependent on pixel connectedness;

creating bounding area definitions each corresponding to a character using direction information and dependent on similar edge pixel labels;

filtering bounding area definitions using direction information to identify character foreground and background information; and

merging filtered adjacent bounding area definitions to form a line definition dependent on the adjacent bounding area definitions relative location to each other.

16. The computer readable medium as described in claim 15 wherein the computer performs labeling edge pixels dependent on identifying end point edge pixels.

17. The computer readable medium as described in claim 15 wherein filtering using direction information is performed by scanning bounding area definitions with a projection line to identify pairs of edge pixels and associated gradient vectors of each pair of edge pixels to evaluate associated foreground and background information of the pair.

18. The computer readable medium as described in claim 15 wherein filtering using direction information is performed by statistically analyzing the gradient vector information for the pairs of edge pixels to identify a foreground shade associated with each bounding area definition.

19. The computer readable medium as described in claim 15 wherein filtering is performed by analyzing geometric characteristics of the bounding area definitions.

20. The computer readable medium as described in claim 15 wherein filtering is performed by identifying edge pixel pairs from direction information to identify the character background and foreground information.

21. The computer readable medium as described in claim 15 wherein combining adjacent bounding area definitions to form a line definition is performed by:

a) assigning a current bounding area definition to a current line definition;

b) searching a pre-defined area in the proximity of the current line;

c) locating a new bounding area at least overlapping the pre-defined area;

d) merging the definition of the new bounding area into the current line definition;

e) repeating b)-d) until no more new bounding areas.