Automatic identification of fields and labels in forms

• US 9,058,516 B2
• Filed: 03/08/2012
• Issued: 06/16/2015
• Est. Priority Date: 03/07/2012
• Status: Active Grant

• Current Assignee: Ricoh Company Limited
• Original Assignee: Ricoh Company Limited
• Inventors: Barrus, John W.
• Primary Examiner(s): Debrow, James J
• Application Number:
  US13/415,766
• Publication Number:
  US 20130238966A1
• Time in Patent Office: 1,195 Days
• Field of Search: 715/233
• US Class Current: 1/1
• CPC Class Codes: G06V 30/412 Layout analysis of documents structured with printed lines or input boxes, e.g. business forms or tables

Abstract

A system and method for processing form images including strokes. A controller receives a plurality of form images including a plurality of strokes. A stroke identification module identifies the position of each stroke in each of the form images. A geometry engine generates an overlay of the plurality of form images and identifies a group of overlapping strokes from the overlay. The geometry engine generates a field bounding box encompassing the group of strokes, the field bounding box representing a field in the plurality of form images. The geometry engine crops a field image from each form image based on the size and position of the field bounding box. A label detector analyzes an area around the field image in the form image to determine a label and generates a label image.

Claims

1. A computer-implemented method for generating symbolic information for a first set of field images associated with a first field, the method comprising:

receiving the first set of field images associated with the first field, the first set of field images associated with the first field cropped from a plurality of form images;

retrieving a first label image associated with the first field, the first label image associated with the first field cropped from one of the plurality of form images;

determining a match for the first label image from a classification dictionary;

associating, with one or more processors, symbolic information corresponding to the match for the first label image with the first label image;

determining a subject matter associated with the first label image using the symbolic information associated with the first label image;

identifying a subset of the classification dictionary using the subject matter associated with the first label image;

determining a match for each field image of the first set of field images from the subset of the classification dictionary; and

associating symbolic information corresponding to the match for each field image with each corresponding field image of the first set of field images.

2. The method of claim 1, wherein the first label image is a master label image.

3. The method of claim 1, wherein the subset of the classification dictionary limits a vocabulary for determining a match to numbers, names, abbreviations of months and symbols.

4. The method of claim 1, further comprising:

generating a new image based on the symbolic information of the first label image; and

indexing the new image with the first label image and the first set of field images.

5. The method of claim 1, further comprising:

determining a label match for the symbolic information of the first label image;

generating a new image with the label match; and

indexing the label match with the first label image and the first set of field images.

6. The method of claim 1, further comprising:

receiving a second set of field images associated with a second field;

determining a geometric relationship between a first position of the first set of field images and a second position of the second set of field images; and

associating the first field and the second field based on the geometric relationship.

7. The method of claim 6, further comprising:

analyzing the first label image and a second label image associated with the second set of field images; and

associating the first field with the second field based on the analysis of the first label image and the second label image.

8. The method of claim 6, further comprising:

analyzing a pattern of stroke usage between the first set of field images associated with the first field and the second set of field images associated with the second field; and

designating the first field and the second field as mutually exclusive based on the analysis of the pattern of stroke usage.

9. A system for generating symbolic information for a first set of field images associated with a first field, the system comprising:

a processor;

a table generator stored on a memory and executable by the processor, the table generator for receiving the first set of field images associated with the first field, the first set of field images associated with the first field cropped from a plurality of form images; and

a symbolic representation module coupled to the table generator, the symbolic representation module for retrieving a first label image associated with the first field, the first label image associated with the first field cropped from one of the plurality of form images, determining a match for the first label image from a classification dictionary, associating symbolic information corresponding to the match for the first label image with the first label image, determining a subject matter associated with the first label image using the symbolic information associated with the first label image, identifying a subset of the classification dictionary using the subject matter associated with the first label image, determining a match for each field image of the first set of field images from the subset of the classification dictionary and associating symbolic information corresponding to the match for each field image with each corresponding field image of the first set of field images.

10. The system of claim 9, wherein the first label image is a master label image.

11. The system of claim 9, wherein the subset of the classification dictionary limits a vocabulary for determining a match to numbers, names, abbreviations of months and symbols.

12. The system of claim 9, wherein the symbolic representation module generates a new image based on the symbolic information of the first label image.

13. The system of claim 9, wherein the symbolic representation module determines a label match for the symbolic information of the first label image and generates a new image with the label match.

14. The system of claim 9, wherein the table generator receives a second set of field images associated with a second field, determines a geometric relationship between a first position of the first set of field images and a second position of the second set of field images and associates the first field and the second field based on the geometric relationship.

15. The system of claim 14, wherein the table generator analyzes the first label image and a second label image associated with the second set of field images and associates the first field with the second field based on the analysis of the first label image and the second label image.

16. The system of claim 14, wherein the table generator analyzes a pattern of stroke usage between the first set of field images associated with the first field and the second set of field images associated with the second field and designates the first field and the second field as mutually exclusive based on the analysis of the pattern of stroke usage.

17. A computer program product stored on a non-transitory computer readable medium including a computer readable program, wherein the computer readable program when executed on a computer causes the computer to:

receive a first set of field images associated with a first field, the first set of field images associated with the first field cropped from a plurality of form images;

retrieve a first label image associated with the first field, the first label image associated with the first field cropped from one of the plurality of form images;

determine a match for the first label image from a classification dictionary;

associate symbolic information corresponding to the match for the first label image with the first label image;

determine a subject matter associated with the first label image using the symbolic information associated with the first label image;

identify a subset of the classification dictionary using the subject matter associated with the first label image;

determine a match for each field image of the first set of field images from the subset of the classification dictionary; and

associate symbolic information corresponding to the match for each field image with each corresponding field image of the first set of field images.

18. The computer program product of claim 17, wherein the first label image is a master label image.

19. The computer program product of claim 17, wherein the subset of the classification dictionary limits a vocabulary for determining a match to numbers, names, abbreviations of months and symbols.

20. The computer program product of claim 17, further causing the computer to:

generate a new image based on the symbolic information of the first label image; and

index the new image with the first label image and the first set of field images.