Character segmentation and recognition method

• US 10,430,681 B2
• Filed: 06/26/2015
• Issued: 10/01/2019
• Est. Priority Date: 02/13/2015
• Status: Active Grant

• Current Assignee: GRG Banking Equipment Company Limited (City of Guangzhou)
• Original Assignee: GRG Banking Equipment Company Limited (City of Guangzhou)
• Inventors: Wang, Weifeng, Qiu, Xinhua, Wang, Rongqiu, Wang, Kun
• Primary Examiner(s): Mariam, Daniel G
• Application Number:
  US15/549,126
• Publication Number:
  US 20180096216A1
• Time in Patent Office: 1,558 Days
• Field of Search: 382173, 382177, 382209
• CPC Class Codes: G06V 30/153 using recognition of characters or words
• CPC Class Codes: G06V 30/158 using character size, text spacings or pitch estimation

Abstract

Provided is a character segmentation and recognition method. The method includes: collecting image data to obtain a to-be-recognized image; positioning a character line candidate region on the to-be-recognized image; obtaining pre-set character line prior information, where the character line prior information includes the number of characters, character spacing and a character size; obtaining a corresponding segmentation point template based on the character line prior information; obtaining credible degrees of different positions on the character line candidate region traversed by the segmentation point template; determining a position with the highest credible degree as an optimal segmentation position; segmenting the character line candidate region based on the segmentation point template and the optimal segmentation position to obtain multiple single character regions; and performing character recognition on each of the single character regions to obtain a corresponding recognition result.

Claims

1. A character segmentation and recognition method, comprising:

collecting image data to obtain a to-be-recognized image;

positioning a character line candidate region on the to-be-recognized image, wherein the character line candidate region comprises a plurality of characters which do not overlap with each other;

obtaining pre-set character line prior information, wherein the character line prior information comprises a quantity of the plurality of characters, a spacing among the plurality of characters, and a size of the plurality of characters;

obtaining a segmentation point template based on the character line prior information, wherein the segmentation point template comprises boundary frames of each of the plurality of characters;

traversing the segmentation point template within the character line candidate region, to obtain credible degrees of different positions in the character line candidate region;

determining an optimal segmentation position from the different positions, wherein the credible degree of the optimal segmentation position is largest among the credible degrees of the different positions;

segmenting the character line candidate region based on the segmentation point template and the optimal segmentation position to obtain a plurality of single-character regions which do not overlap with each other, wherein each of the plurality of single-character regions comprises one of the plurality of characters; and

performing character recognition on each of the plurality of single-character regions to obtain a corresponding recognition result;

wherein traversing the segmentation point template within the character line candidate region, to obtain credible degrees of different positions in the character line candidate region comprises;

recording position information of the segmentation point template, in response to the segmentation point being traversed to each of the different positions; and

obtaining the credible degrees corresponding to the position information; and

wherein the obtaining the credible degrees corresponding to the position information comprises;

obtaining the first number of first effective pixel points of the character line candidate region on a left boundary of a character segmentation point, wherein the segmentation point template comprises a plurality of character segmentation points, each of the character segmentation points corresponds to one character on a character line, each of the character segmentation points comprises the left boundary and a right boundary, and the first effective pixel points are pixel points with gray values in a pre-set first threshold range;

obtaining a first evaluation value based on the first number and a pre-set first weight;

obtaining the second number of second effective pixel points of the character line candidate region on the right boundary of the character segmentation point, wherein the second effective pixel points are pixel points with gray values in a pre-set second threshold range;

obtaining a second evaluation value based on the second number and a pre-set second weight;

obtaining the third number of third effective pixel points of the character line candidate region within the segmentation point template, wherein the third effective pixel points are pixel points with gray values in a pre-set third threshold range;

obtaining a third evaluation value based on the third number and a pre-set third weight; and

obtaining the credible degrees corresponding to the position information based on the first evaluation value, the second evaluation value and the third evaluation value.

2. The method according to claim 1, wherein the obtaining the pre-set character line prior information comprises:

obtaining the pre-set character line prior information based on a type of the to-be-recognized image;

or

obtaining the pre-set character line prior information based on an orientation feature of the character line candidate region.

3. The method according to claim 2, wherein after obtaining the corresponding segmentation point template based on the character line prior information, the method further comprises:

obtaining a binary image of the to-be-recognized image;

obtaining a projection waveform of the binary image in a horizontal direction; and

updating an upper boundary and a lower boundary of the segmentation point template based on the projection waveform.

4. The method according to claim 2, wherein the positioning the character line candidate region on the to-be-recognized image comprises:

positioning a region of interest where a character line locates on the to-be-recognized image;

performing affine transformation and bilinear interpolation on the region of interest to obtain a correction image; and

positioning the character line candidate region on the correction image.

5. The method according to claim 2, wherein the performing character recognition on each of the single character regions to obtain the corresponding recognition result comprises:

obtaining character feature data in the single character region; and

performing character recognition based on the character feature data to obtain the recognition result.

6. The method according to claim 1, wherein the obtaining the corresponding segmentation point template based on the character line prior information comprises:

creating and initializing the segmentation point template based on the character line prior information;

or

obtaining the corresponding segmentation point template based on the character line prior information and a pre-set segmentation index table, wherein the segmentation index table comprises a correspondence between the character line prior information and the segmentation point template.

7. The method according to claim 6, wherein after obtaining the corresponding segmentation point template based on the character line prior information, the method further comprises:

obtaining a binary image of the to-be-recognized image;

obtaining a projection waveform of the binary image in a horizontal direction; and

updating an upper boundary and a lower boundary of the segmentation point template based on the projection waveform.

8. The method according to claim 6, wherein the positioning the character line candidate region on the to-be-recognized image comprises:

positioning a region of interest where a character line locates on the to-be-recognized image;

performing affine transformation and bilinear interpolation on the region of interest to obtain a correction image; and

positioning the character line candidate region on the correction image.

9. The method according to claim 6, wherein the performing character recognition on each of the single character regions to obtain the corresponding recognition result comprises:

obtaining character feature data in the single character region; and

performing character recognition based on the character feature data to obtain the recognition result.

10. The method according to claim 1, wherein after obtaining the corresponding segmentation point template based on the character line prior information, the method further comprises:

obtaining a binary image of the to-be-recognized image;

obtaining a projection waveform of the binary image in a horizontal direction; and

updating an upper boundary and a lower boundary of the segmentation point template based on the projection waveform.

11. The method according to claim 10, wherein the updating the upper boundary and the lower boundary of the segmentation point template based on the projection waveform comprises:

determining and updating the uppermost position at which a grey value is greater than a pre-set fourth threshold in the projection waveform as the upper boundary of the segmentation point template; and

determining and updating the lowermost position at which a grey value is greater than a pre-set fifth threshold in the projection waveform as the lower boundary of the segmentation point template.

12. The method according to claim 1, wherein the positioning the character line candidate region on the to-be-recognized image comprises:

positioning a region of interest where a character line locates on the to-be-recognized image;

performing affine transformation and bilinear interpolation on the region of interest to obtain a correction image; and

positioning the character line candidate region on the correction image.

13. The method according to claim 1, wherein the performing character recognition on each of the single character regions to obtain the corresponding recognition result comprises:

obtaining character feature data in the single character region; and

performing character recognition based on the character feature data to obtain the recognition result.

14. The method according to claim 13, wherein the character feature data is character contour feature data.