Segmentation process for machine reading of handwritten information

• US 5,101,439 A
• Filed: 08/31/1990
• Issued: 03/31/1992
• Est. Priority Date: 08/31/1990
• Status: Expired due to Term

• Current Assignee: NCR Corporation
• Original Assignee: AT&T, Inc.
• Inventors: Kiang, Richard K.
• Primary Examiner(s): Razavi, Michael
• Application Number:
  US07/576,246
• Time in Patent Office: 578 Days
• Field of Search: 382/9, 382/13, 382/14, 382/15, 382/39, 382/46, 382/16, 382/18, 382/30, 382/37, 382/44, 382/22, 382/10, 358/107, 358/101
• US Class Current: 382/174
• CPC Class Codes: G06V 30/10 Character recognition
• CPC Class Codes: G06V 30/166 Normalisation of pattern dimensions

Abstract

An improved machine process for recognizing the transition point between letters and digits in handwritten text is disclosed. The script, digits and/or letters, is mapped onto a pixel matrix. The entropy for each column for a range of assumed slant angles is analyzed to determine the angle at which image entropy is a minimum. By reverse-shifting the pixel array an amount equal to this angle, the slant component of the script is verticalized and the segmentation step which follows is more accurately accomplished.

Claims

1. In the machine-reading of original handwritten word or number script, a method for normalizing the characteristic slant angle in the script, comprising the steps of:

generating a pixel map of said original script, with said script image aligned on said map in a direction substantially parallel to the horizontal axis of said pixel map;

said map comprising vertical columns and horizontal rows of dark and white pixels, said dark pixels displaying the script of said word or number, and said display exhibiting said characteristic script slant angle relative to said vertical columns;

moving said rows of pixels thereby to generate a first altered script image having a first slant angle;

calculating the total number of dark pixels in said first altered image;

calculating the number of dark pixels in each said vertical column of said first altered image;

calculating the proportion of dark pixels in each column relative to said total dark pixels in said first altered image;

using said proportionality value, computing the entropy value of said first altered image for said first slant angle;

storing said computed entropy value and its associated first slant angle;

successively repositioning said rows of pixels thereby to generate a succession of further altered script images, each of said last-named images having a different slant angle;

computing and storing the entropy values and the respective associated slant angle values for each said further altered image;

determining the angular difference between that slant angle associated with the minimum entropy value and the characteristic slant angle; and

with the pixel rows in their original image position, shifting said pixel rows by the amount of said angular difference, thereby to create an image of said handwritten script that has its characteristic slant angle removed.

2. The method in accordance with claim 1, wherein said different slant angles comprise a range with end points located on either side of said characteristic slant angle.

3. The method in accordance with claim 1, wherein said successive repositioning of said rows of pixels comprises moving each individual pixel row an amount linerally proportional to its distance from a selected stationary pixel reference row.

4. The method in accordance with claim 1, wherein said successive repositioning of said rows of pixels comprises keeping the top-most pixel row stationary, shifting the bottom-most pixel row to the right in incremental steps substantially equal to the length of a pixel side, and shifting the intermediary rows by amounts proportional to their respective distances from the reference row.

5. The method in accordance with claim 4, comprising the further step of segmenting the now-normalized script into its discrete letters or digits.

6. Apparatus for normalizing the characteristic slant angle in original handwritten word or number script, comprising:

means for generating a pixel map of said script, with said script image aligned on said map in a direction substantially parallel to the horizontal axis of said pixel map;

said map comprising vertical columns and horizontal rows of dark and white pixels, said dark pixels displaying the script of said word or number, and said display exhibiting said characteristic script slant angle relative to said vertical columns;

means for moving said rows of pixels thereby to generate a first altered script image having a first slant angle;

means for calculating the total number of dark pixels in said first altered image;

means for calculating the number of dark pixels in each said vertical column of said first altered image;

means for calculating the proportion of dark pixels in each column relative to said total dark pixels in said first altered image;

means for computing the entropy value of said first altered image;

means for storing said computed entropy value and its associated first slant angle;

means for successively repositioning said rows of pixels thereby to generate a succession of further altered script images, each of said last-named images having a different slant angle;

means for computing and storing the entropy values and the respective associated slant angle values for each said further altered image;

means for determining the angular difference between that slant angle associated with the minimum entropy value and the characteristic slant angle; and

with the pixel rows in their original image position, means for shifting said pixel rows by the amount of said angular difference, thereby to create an image of said handwritten script that has its characteristic slant angle removed.

7. Apparatus in accordance with claim 6, further comprising means for segmenting the now-normalized script into its discrete letters or digits.