Form recognition using reference areas

• US 6,778,703 B1
• Filed: 04/19/2000
• Issued: 08/17/2004
• Est. Priority Date: 04/19/2000
• Status: Expired due to Term

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Zlotnick, Aviad
• Primary Examiner(s): CHANG, JON CARLTON
• Application Number:
  US09/553,122
• Time in Patent Office: 1,581 Days
• Field of Search: 382/209, 382/217, 382/218, 382/219, 382/220
• US Class Current: 382/218
• 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 method for processing an input image includes automatically selecting reference areas on each of a plurality of different templates. Reference areas on the input image are compared to the selected reference areas on at least some of the templates, so as to identify the template that most closely matches the input image, and information is extracted from the input image based on the identified template. Typically, the input image is an image of a form document, having fields defined by one of the templates, which fields are filled in with the information. The input image is registered with the identified template in order to extract the information from the fields.

Claims

1. A method for processing an input image, comprising:

for each of a plurality of different templates, computing a reference metric for each of a multiplicity of candidate areas;

selecting reference areas from among the candidate areas on each of the plurality of different templates responsive to the reference metrics thereof;

comparing reference areas on the input image to the selected reference areas on at least some of the templates, so as to compute a matching score for each of the templates indicating a degree of similarity between each template and the input image;

identifying the template whose matching score indicates the greatest degree of similarity; and

extracting information from the input image based on the identified template.

2. A method according to claim 1, wherein the input image comprises an image of a form document, having fields defined by one of the templates, which fields are filled in with the information, and wherein extracting the information comprises registering the input image with the identified template.

3. A method according to claim 2, wherein computing the reference metric for each of the candidate areas comprises estimating a likelihood that the candidate area will be filled in.

4. A method according to claim 3, wherein selecting the reference areas comprises rejecting areas that comprise runs of white pixels of a length exceeding a predetermined criterion.

5. A method according to claim 1, wherein computing the reference metric for each of the candidate areas comprises estimating a likelihood of a change in appearance of the area due to contrast variations in scanning an object that appears in the input image.

6. A method according to claim 5, wherein selecting the areas comprises rejecting areas that comprise predetermined concentrations of black pixels.

7. A method according to claim 1, wherein computing the reference metric comprises determining a risk score indicative of a likelihood of confusion in identifying the template based on a given reference area, and wherein selecting the reference areas comprises selecting one or more reference areas having respective risk scores within a predetermined limit.

8. A method according to claim 1, wherein selecting the reference areas comprises selecting at least one of the candidate areas in each of a number of predetermined sectors of each template.

9. Apparatus for processing an input image, comprising:

a memory, which is adapted to store images of a plurality of different templates; and

an image processor, which is adapted to compute a reference metric for each of a multiplicity of candidate areas on each of the templates and to select reference areas from among the candidate areas on each of the plurality of different templates responsive to the reference metrics thereof, and further to receive the input image and compare reference areas on the input image to the selected reference areas on at least some of the templates, so as to compute a matching score for each of the templates indicating a degree of similarity between each template and the input image and to identify the template whose matching score indicates the greatest degree of similarity, and extract information from the input image based on the identified template.

10. Apparatus according to claim 9, wherein the input image comprises an image of a form document, having fields defined by one of the templates, which fields are filled in with the information, and wherein the processor is adapted to register the input image with the identified template.

11. Apparatus according to claim 10, and comprising an image input device, which is adapted to capture the image of the form document and to convey the image to the processor.

12. A computer software product for processing an input image, comprising a computer-readable medium having program instructions stored therein, which instructions, when read by a computer, cause the computer to compute a reference metric for each of a multiplicity of candidate areas reference areas on each of a plurality of different templates and to select reference areas from among the candidate areas on each of the plurality of different templates responsive to the reference metrics thereof, and further to compare reference areas on the input image to the selected reference areas on at least some of the templates, so as to compute a matching score for each of the templates indicating a degree of similarity between each template and the input image and to identify the template whose matching score indicates the greatest degree of similarity, and to extract information from the input image based on the identified template.