Image processing method for character recognition, character recognition apparatus using this method, and program

• US 9,704,052 B2
• Filed: 11/28/2012
• Issued: 07/11/2017
• Est. Priority Date: 03/05/2012
• Status: Active Grant

• Current Assignee: Omron Corporation
• Original Assignee: Omron Corporation
• Inventors: Fujieda, Shiro
• Primary Examiner(s): Chu, Randolph I
• Application Number:
  US14/379,118
• Publication Number:
  US 20150023593A1
• Time in Patent Office: 1,686 Days
• Field of Search: None
• CPC Class Codes: G06V 30/10 Character recognition
• CPC Class Codes: G06V 30/1478 of characters or characters lines
• CPC Class Codes: G06V 30/18095 Summing image-intensity values; Projection and histogram analysis
• CPC Class Codes: G06V 30/40 Document-oriented image-based pattern recognition

Abstract

An image processing method of processing a gray-scale image of character strings arranged in a plurality of lines of text, for character recognition, includes setting at least a part of a range, in which the character strings in the gray-scale image area are distributed, as a search area, and setting, for each of a plurality of points arranged along one of two axes, representing coordinates in the image, corresponding to an arrangement of the line of text, a plurality of lines extending across the search area, with the point as a starting point, and accumulating, for each of the lines, pixel data in the line within the search area.

Claims

1. An image processing method of processing a gray-scale image of character strings arranged in a plurality of lines of text, for character recognition, the method comprising:

setting at least a part of a range of a target area, in which the character strings in the gray-scale image area are distributed, as a search area;

setting, for each of a plurality of points arranged along one of two axes, representing coordinates in the image, corresponding to an arrangement of the line of text, a plurality of lines extending across the search area, with the point as a starting point, and accumulating, for each of the lines, pixel data in the line within the search area;

obtaining, for each point as the starting point of the line, a maximum value of accumulated values in a group of lines set for the point, and specifying at least one local maximum value in a distributed pattern generated by arranging the maximum values obtained for the points, in accordance with the arrangement of the points; and

detecting each line of text of the character strings in the image based on a line, in which the accumulated value corresponding to the specified local maximum value is obtained, set as a line passing between the character strings,

wherein the search area is an inner area of the target area offset by a first predetermined distance above the search area and a second predetermined distance below the search area,

wherein an area above the first predetermined distance and an area below the second predetermined distance are blank space.

3. The image processing method for character recognition according to claim 1, further comprising:

performing, for each character string detected by the line passing through the character strings;

projecting an image of the character string onto one of the two axes, representing the coordinates in the image, not corresponding to the arrangement of the line of text, and

individually detecting a character in the character string, based on a result of the projection.

4. The image processing method for character recognition according to claim 1, further comprising:

performing, for each character string detected by the line passing through the character strings;

rotationally correcting an image of the character string, in such a manner that the line used for the detection extends along one of the two axes, representing the coordinates in the image, not corresponding to the arrangement of the line of text,

projecting the corrected image onto the one of two axes not corresponding to the arrangement of the line of text, and

individually detecting a character in the character string based on a result of the projection.

5. The image processing method for character recognition according to claim 1,

wherein, in the detecting each line of text of the character strings in the image, an image as a processing target is rotationally corrected in such a manner that the line passing between the character strings extends along one of the two axes, representing the coordinates in the image, not corresponding to the arrangement of the line of text, and then each line of text of the character strings is detected based on the line passing between the character strings in the corrected image, and

wherein, for each character string detected by the line passing through the character strings, projecting an image of the character string onto the one of two axes not corresponding to the arrangement of the line of text, and individually detecting a character in the character string based on a result of the projection.

10. The image processing method for character recognition according to claim 1, further comprising:

performing, for each character string detected by the line passing through the character strings;

projecting an image of the character string onto one of the two axes, representing the coordinates in the image, not corresponding to the arrangement of the line of text, and

individually detecting a character in the character string, based on a result of the projection.

2. An image processing method of processing a gray-scale image of character strings arranged in a plurality of lines of text, for character recognition, the method comprising:

setting at least a part of a range of a target area, in which the character strings in the gray-scale image area are distributed, as a search area;

setting, for each of a plurality of points arranged along one of two axes, representing coordinates in the image, corresponding to an arrangement of the line of text, a plurality of lines extending from the point and across the search area, and accumulating, for each of the lines, pixel data in the line within the search area;

obtaining, for each point as a starting point of the line, a minimum value of accumulated values in a group of lines set for the point, and specifying at least one local minimum value in a distributed pattern generated by arranging the minimum values obtained for the points, in accordance with the arrangement of the points; and

detecting each line of text of the character strings in the image based on a line, in which the accumulated value corresponding to the specified local minimum value is obtained, set as a line passing between the character strings,

wherein the search area is an inner area of the target area offset by a first predetermined distance above the search area and a second predetermined distance below the search area,

wherein an area above the first predetermined distance and an area below the second predetermined distance are blank space.

11. The image processing method for character recognition according to claim 2, further comprising:

performing, for each character string detected by the line passing through the character strings;

rotationally correcting an image of the character string, in such a manner that the line used for the detection extends along one of the two axes, representing the coordinates in the image, not corresponding to the arrangement of the line of text,

projecting the corrected image onto the one of two axes not corresponding to the arrangement of the line of text, and

individually detecting a character in the character string based on a result of the projection.

12. The image processing method for character recognition according to claim 2,

wherein, in the detecting each line of text of the character strings in the image, an image as a processing target is rotationally corrected in such a manner that the line passing between the character strings extends along one of the two axes, representing the coordinates in the image, not corresponding to the arrangement of the line of text, and then each line of text of the character strings is detected based on the line passing between the character strings in the corrected image, and

wherein, for each character string detected by the line passing through the character strings, projecting an image of the character string onto the one of two axes not corresponding to the arrangement of the line of text, and individually detecting a character in the character string based on a result of the projection.

6. A character recognition apparatus configured to input a gray-scale image generated by capturing an image of a character string, as a recognition target, cut out each character in the character string in the image, and then match each cut out character against models registered in advance to recognize the each character, the apparatus comprising:

an area setting unit that sets a search area from a target area in the gray-scale image as the recognition target;

an accumulator that sets, for each of a plurality of points arranged along one of two axes representing coordinates in the image, a plurality of lines extending across the search area with the point as a starting point, and accumulates, for each of the lines, pixel data in the line within the search area;

a specification unit that obtains, for each point as the starting point of the line, a maximum value of accumulated values in a group of lines set for the point, and specifies a local maximum value in a distributed pattern generated by arranging the maximum values obtained for the points in accordance with the arrangement of the points;

a character string detector that detects each line of text of the character strings in the image based on a line, in which the accumulated value corresponding to the local maximum value specified by the specification unit is obtained; and

a character recognition unit that individually cuts out, for each character string detected by the character string detector, a character in the character string, and performs matching processing against the models,

wherein the search area is an inner area of the target area offset by a first predetermined distance above the search area and a second predetermined distance below the search area,

wherein an area above the first predetermined distance and an area below the second predetermined distance are blank space.

7. A character recognition apparatus configured to input a gray-scale image generated by capturing an image of a character string, as a recognition target, cut out each character in the character string in the image, and then match each cut out character against models registered in advance to recognize the each character, the apparatus comprising:

an area setting unit that sets a search area from a target area in the gray-scale image as the recognition target;

an accumulator that sets, for each of a plurality of points arranged along one of two axes representing coordinates in the image, a plurality of lines extending across the search area with the point as a starting point, and accumulates, for each of the lines, pixel data in the line within the search area;

a specification unit that obtains, for each point as the starting point of the line, a minimum value of accumulated values in a group of lines set for the point, and specifies a local minimum value in a distributed pattern generated by arranging the minimum values obtained for the points in accordance with the arrangement of the points;

a character string detector that detects each line of text of the character strings in the image based on a line, in which the accumulated value corresponding to the local minimum value specified by the specification unit is obtained; and

a character recognition unit that individually cuts out, for each character string detected by the character string detector, a character in the character string, and perform matching processing against the models,

wherein the search area is an inner area of the target area offset by a first predetermined distance above the search area and a second predetermined distance below the search area,

wherein an area above the first predetermined distance and an area below the second predetermined distance are blank space.

8. A program stored on a non-transitory computer readable medium for causing a computer to operate as a character recognition apparatus configured to input a gray-scale image generated by capturing an image of a character string, as a recognition target, cut out each character in the character string in the image, and then matching each cut out character against models registered in advance to recognize the each character, the program causing the computer to perform:

setting a search area from a target area in the gray-scale image as the recognition target;

setting, for each of a plurality of points arranged along one of two axes representing coordinates in the image, a plurality of lines extending across the search area with the point as a starting point;

accumulating, for each of the lines, pixel data in the line within the search area;

obtaining, for each point as the starting point of the line, a maximum value of accumulated values in a group of lines set for the point;

specifying a local maximum value in a distributed pattern generated by arranging the maximum values obtained for the points in accordance with the arrangement of the points;

detecting each line of text of the character strings in the image based on a line, in which the accumulated value corresponding to the specified local maximum value is obtained;

individually cutting out, for each detected character string, a character in the character string; and

performing matching processing against the models,

wherein the search area is an inner area of the target area offset by a first predetermined distance above the search area and a second predetermined distance below the search area,

wherein an area above the first predetermined distance and an area below the second predetermined distance are blank space.

9. A program stored on a non-transitory computer readable medium for causing a computer to operate as a character recognition apparatus configured to input a gray-scale image generated by capturing an image of a character string, as a recognition target, cut out each character in the character string in the image, and then matching each of cut out character against models registered in advance to recognize the each character, the program causing the computer to perform:

setting a search area from a target area in the gray-scale image as the recognition target;

setting, for each of a plurality of points arranged along one of two axes representing coordinates in the image, a plurality of lines extending across the search area with the point as a starting point;

accumulating, for each of the lines, pixel data in the line within the search area;

obtaining, for each point as the starting point of the line, a minimum value of accumulated values in a group of lines set for the point;

specifying a local minimum value in a distributed pattern generated by arranging the minimum values obtained for the points in accordance with the arrangement of the points;

detecting each line of text of the character strings in the image based on a line, in which the accumulated value corresponding to the specified local minimum value is obtained;

individually cutting out, for each detected character string, a character in the character string; and

performing matching processing against the models,

wherein the search area is an inner area of the target area offset by a first predetermined distance above the search area and a second predetermined distance below the search area,

wherein an area above the first predetermined distance and an area below the second predetermined distance are blank space.