Method of distinguishing handwritten and machine-printed images

• US 7,072,514 B1
• Filed: 02/06/2003
• Issued: 07/04/2006
• Est. Priority Date: 02/06/2003
• Status: Active Grant

• Current Assignee: National Security Agency
• Original Assignee: The United States of America As Represented By The Director National Security Agency
• Inventors: Thouin, Paul D.
• Primary Examiner(s): Bella, Matthew C.
• Assistant Examiner(s): DESIRE, GREGORY M
• Application Number:
  US10/361,036
• Time in Patent Office: 1,244 Days
• Field of Search: 382/170, 382/171, 382/176, 382/179, 382/187, 382/194, 382/224, 715/511, 715/541
• US Class Current: 382/187
• CPC Class Codes: G06V 30/2455 Discrimination between machine-print, hand-print and cursive writing

Abstract

The present invention is a method of categorizing an image as handwritten, machine-printed, and unknown. First, the image is received. Next, connected components are identified. Next, a bounding box encloses each connected component. Next, a height and width is computed for each bounding box. Next, a sum and maximum horizontal run for each connected component are computed. Next, connected components that are suspected of being characters are identified. If the number of suspected characters is less than or equal to a first user-definable number then the image is categorized as unknown. If the number of suspected characters is greater than the first user-definable number then determine if matches exist amongst the suspected characters. Next, compute a score based on the suspected characters and the number of matches and categorize the image as either handwritten, machine-printed, or unknown.

Claims

1. A method of categorizing an image as handwritten, machine-printed, and unknown,

comprising the steps of:

(a) receiving an image;

(b) identifying connected components within the image;

(c) enclosing each connected component within a bounding box;

(d) computing a height and a width of each bounding box;

(e) computing a sum and maximum horizontal run for each connected component, where the sum is the sum of all pixels in the corresponding connected component, and where the maximum horizontal run is the longest consecutive number of horizontal pixels in the corresponding connected component;

(f) identifying connected components that are suspected of being characters;

(g) if the number of suspected characters is less than or equal to a first user-definable number then categorizing the image as unknown and stopping, otherwise, proceeding to the next step;

(h) if the number of suspected characters is greater than the first user-definable number then comparing the suspected characters to determine if matches exist, where a match exists between a pair of suspected characters if the suspected characters in the pair have the same height and width, if each suspected character in the pair has a height that is less than 4 times its width, and if each suspected character in the pair has a width that is less than 4 times its height; and

(i) computing a score based on the suspected characters and the number of matches and categorizing the image into one of a group of categories consisting of handwritten, machine-printed, and unknown.

2. The method of claim 1, wherein the first user-definable number is 30.

3. The method of claim 1, wherein the step of identifying connected components that are suspected of being characters is comprised of the steps of:

(a) discarding each connected component that has a sum less than a second user-definable number;

(b) discarding each connected component that has a sum greater than a third user-definable number;

(c) creating a histogram of the sums of the connected components that were not discarded;

(d) identifying as suspected characters the connected components within a first user-definable range within the histogram;

(e) if there are less than 30 suspected characters and the first user-definable range cannot be expanded then categorizing the image as unknown and stopping, otherwise proceeding to the next step;

(f) if there are less than 30 suspected characters and the first user-definable range can be expanded then expanding the first user-definable range and returning to the fourth step, otherwise, proceeding to the next step; and

(g) if there are more than 30 suspected characters identified then proceeding to step (h) in claim 1.

4. The method of claim 3, wherein the second user-definable number is 50.

5. The method of claim 3, wherein the third user-definable number is 5000.

6. The method of claim 3, wherein the first user-definable range within the histogram is from a 65^th percentile to a 75^th percentile.

7. The method of claim 3, wherein the first user-definable range is expanded by 10 percentiles.

8. The method of claim 1, wherein the step of computing a score based on the suspected characters and the number of matches and categorizing the image into one of a group of categories consisting of handwritten, machine-printed, and unknown is comprised of the steps of:

(a) computing a match score, where the match score is the number of suspected characters that match at least one other suspected character divided by the total number of suspected characters;

(b) identifying each suspected character having a maximum horizontal run greater than a fourth user-definable number;

(c) computing, for each suspected character identified in step (b), a ratio of the maximum horizontal run of the suspected character divided by the width of the suspected character;

(d) summing the ratios resulting from step (c);

(e) forming a maximum run ratio as the result of step (d) divided by the total number of ratios computed in step (c);

(f) if the match score is greater than 0.5 then categorizing the image as machine-printed and stopping, otherwise, proceeding to the next step;

(g) if the maximum run ratio is greater than 0.8 then categorizing the image as machine-printed and stopping, otherwise, proceeding to the next step;

(h) if the maximum run ratio is less than 0.5 then categorizing the image as handwritten and stopping, otherwise, proceeding to the next step;

(i) if the maximum run ratio is greater than 0.7 and the match score is greater than 0.05 then categorizing the image as machine-printed and stopping, otherwise, proceeding to the next step;

(j) if the match score is less than 0.1 then categorizing the image as handwritten and stopping, otherwise, proceeding to the next step;

(k) if the match score is greater than 0.4 then categorizing the image as machine-printed and stopping, otherwise, proceeding to the next step; and

(l) if the image has been rotated then categorizing the image as unknown and stopping, otherwise, rotating the image 90 degrees and returning to step (b) in claim 1.

9. The method of claim 8, wherein the fourth user-definable number is 10.