Handwritten word recognition based on geometric decomposition

• US 8,077,973 B2
• Filed: 01/30/2006
• Issued: 12/13/2011
• Est. Priority Date: 01/28/2005
• Status: Active Grant

• Current Assignee: IMDS Software, Inc.
• Original Assignee: IMDS Software, Inc.
• Inventors: Dong, Jianxiong
• Primary Examiner(s): Werner, Brian
• Assistant Examiner(s): Patel, Nirav G
• Application Number:
  US11/344,505
• Publication Number:
  US 20060193518A1
• Time in Patent Office: 2,143 Days
• Field of Search: 382/186, 382/187, 382/199
• US Class Current: 382/186
• CPC Class Codes: G06V 2201/133 Survey article
• CPC Class Codes: G06V 30/10 Character recognition
• CPC Class Codes: G06V 30/1478 of characters or characters lines
• CPC Class Codes: G06V 30/182 by coding the contour of the pattern
• CPC Class Codes: G06V 30/2264 using word shape

Abstract

A method of recognizing a handwritten word of cursive script includes providing a template of previously classified words, and optically reading a handwritten word so as to form an image representation thereof comprising a bit map of pixels. The external pixel contour of the bit map is extracted and the vertical peak and minima pixel extrema on upper and lower zones respectively of this external contour are detected. Feature vectors of the vertical peak and minima pixel extrema are determined and compared to the template so as to generate a match between the handwritten word and a previously classified word. A method for classifying an image representation of a handwritten word of cursive script is also provided. Also provided is an apparatus for recognizing a handwritten word of cursive script.

Claims

1. A method for classifying an image representation of a handwritten word of cursive script, said method comprising:

optically reading a handwritten word so as to form an image representation thereof comprising a bit map of pixels;

extracting a pixel contour of said bit map;

detecting vertical peak and minima pixel extrema on upper and lower zones of said contour respectively;

detecting local vertical peak pixel extrema on an upper zone of said external contour by determining if a given local pixel is a vertical peak relative to neighbouring pixels;

detecting local vertical minima pixel extrema on a lower zone of said external contour by determining if a given local pixel is a vertical minimum relative to neighbouring pixels;

organizing said peak and minima pixel extrema into respective independent peak and minima sequences comprising extracting features at each said extrema and further classifying extrema into two sequences of extrema on respective said upper and lower zones of said word image pixel contour;

determining the respective feature vectors of said peak and minima sequences; and

classifying said word image according to said peak and minima feature vectors, wherein at least one or more of said features extracted at each said extrema is selected from the group consisting of;

the number of local extrema neighboring a given said extrema on a same closed curve of said word image contour, said local extrema having a convex attribute corresponding to that of said given extrema;

the number of local extrema neighboring a given said extrema on a same closed curve of said word image contour, said local extrema having a different convex attribute from said given extrema;

the lesser of the height difference between a given said extrema and a left neighbouring extrema and of the height difference between said given extrema and a right neighbouring extrema, wherein said left and right neighbouring extrema have convex attribute corresponding to that of said given extrema;

the lesser of the height difference between a given said extrema and a left neighbouring extrema and of the height difference between said given extrema and a right neighbouring extrema, wherein said left and right neighbouring extrema have a different convex attribute than that of said given extrema;

the number of peaks above a said given extrema divided by the total number of peaks on said pixel contour;

the number of peaks below a said given extrema divided by the total number of peaks on said pixel contour;

the y/h position of said given extrema, wherein y represents the y-axis coordinate of said given extrema and h represents the height of said word image;

the lesser of a contour portion length between a given said extrema and a left neighbouring peak and of a contour portion length between a given said extrema and a right neighbouring peak, wherein said neighbouring peaks and said given extrema are on a same closed curve;

the lesser of a contour portion length between a given said extrema and a left neighbouring minima and of a contour portion length between a given said extrema and a right neighbouring minima, wherein said neighbouring minima and said given extrema are on a same closed curve;

the lesser of a height difference between a given said extrema and a left neighbouring peak and of a given said extrema and a and right neighbouring peak, wherein said neighbouring peaks and said given extrema are on a same closed curve;

the lesser of a height difference between a given said extrema and a left neighbouring minima and of a given said extrema and a and right neighbouring minima, wherein said neighbouring minima and said given extrema are on a same closed curve;

the height ratio of a given said extrema and neighboring left and right extrema as defined by

(y_A−

y_tl)/(y_n−

y_tln)

wherein a given said extrema is represented by A, a lowest extrema of said left or right neighbouring extrema is represented by n, y_tl represents the y-coordinate of the top-left corner of a said contour or a said closed curve, y_tln represents the top-left corner of a said contour or a said closed curve where point n is located y_A and y_n represent the y-coordinate of A and n respectfully;

the distance between a given said extrema and a vertical intersection point; and

any combination thereof.

2. A method according to claim 1, wherein said determining comprises determining feature vectors of said vertical peak and minima sequences.

3. A method according to claim 1, wherein said optically reading comprises scanning.

4. A method according to claim 1, wherein a said vertical peak pixel extrema comprises a pixel point on said upper zone of said pixel contour wherein neighbouring pixel segments are convex relative thereto.

5. A method according to claim 1, wherein a said vertical minima pixel extrema comprises a pixel point on said lower zone of said pixel contour wherein neighbouring pixel segments are concave relative thereto.

6. A method according to claim 1, further comprising correcting a slant of said bit map before said extracting of said pixel contour.

7. A method according to claim 6, further comprising correcting a skew of said bit map when correcting said slant.

8. A method according to claim 6, wherein said slant correction comprises

pre-computing a look-up table of the image representation that maps Cartesian coordinates to polar coordinates;

performing a Radon transform in the polar domain of the image representation

calculating a skew angle from the Radon transform.

9. A method according to claim 1, further comprising pruning invalid peak and minima extrema before said determining of said respective feature vectors.

10. A method according to claim 1, wherein said comparing is performed by dynamic time wrapping.

11. A method according to claim 1, wherein said determining said respective feature vectors comprises extracting features at each said extrema selected from the group consisting of:

height difference between a given said extrema and a left neighbouring extrema having a convex attribute corresponding to the that of said given extrema;

height difference between a given said extrema and a right neighbouring extrema having a convex attribute corresponding to the that of said given extrema;

relative horizontal position of given said extrema defined by;