Method and apparatus for recognizing hand-written characters using a weighting dictionary

• US 5,812,697 A
• Filed: 02/14/1997
• Issued: 09/22/1998
• Est. Priority Date: 06/10/1994
• Status: Expired due to Term

• Current Assignee: Foothills IP LLC
• Original Assignee: Nippon Steel Corporation
• Inventors: Ikeda, Yoshitaka, Sakai, Yoshifumi
• Primary Examiner(s): Mancuso, Joseph
• Assistant Examiner(s): Patel, Jayanti K.
• Application Number:
  US08/800,374
• Time in Patent Office: 585 Days
• Field of Search: 382/186, 382/187, 382/220, 382/179, 382/189, 382/225
• US Class Current: 382/186
• CPC Class Codes: G06F 18/28 Determining representative reference patterns, e.g. by averaging or distorting; Generating dictionaries
• CPC Class Codes: G06V 30/373 using a special pattern or subpattern alphabet

Abstract

A method of recognizing handwritten characters. A recognition dictionary is prepared including a plurality of characters and corresponding character clusters. A weighting dictionary is prepared including weighting clusters and corresponding object characters. All segmentation position candidates are selected by segmenting handwritten character ink data. Each of the position candidates is separated in a direction corresponding to a direction in which the handwritten character ink data is inputted from a leading position of the ink data and satisfying a predetermined condition. A character candidate is selected corresponding to each segment of the ink data segmented at the segmentation position candidates by referring to the recognition dictionary and calculating a distance value between each of the segments of the ink data and a selected one of the character clusters. The weighting dictionary is referred to and the distance value calculated for one of the character candidates is weighted if one of the weighting clusters included in the weighting dictionary is similar to any one of the segments of the ink data segmented at the segmentation position candidates and if the object character of the one weighting cluster coincides with the one of the character candidates. A rank of the character candidates is determined in accordance with the weighted distance values thereof.

Claims

1. A method of recognizing handwritten characters, comprising the steps of:

preparing a recognition dictionary including a plurality of characters and corresponding character clusters, and a weighing dictionary including weighing clusters and corresponding object characters;

setting all segmentation position candidates by segmenting handwritten character ink data, each of the position candidates being separated in a direction corresponding to a direction in which the handwritten character ink data is inputted from a leading position of the ink data and satisfying a predetermined condition;

selecting a character candidate corresponding to each segment of the ink data segmented at the segmentation position candidates by referring to the recognition dictionary and calculating a distance value between each of the segments of the ink data and a selected one of the character clusters;

searching the weighing dictionary and applying a weighing quantity to the distance value calculated for one of the character clusters corresponding to the selected character candidate, if a distance between one of the weighing clusters included in the weighing dictionary and any one of the segments of the ink data segmented at the segmentation position candidates is equal to or less than a predetermined threshold value and if the object character of said one weighing cluster coincides with the selected character candidate; and

determining a rank of the character candidates in accordance with the weighted distance values thereof.

2. A method according to claim 1, wherein the weighting of the distance value of the one character candidate is based on a second distance value between the object character of the weighting cluster, which coincides with the one character candidates, and the segment of the ink data corresponding to said one character candidate.

3. A method according to claim 2, wherein the weighting of the distance value becomes larger, as the second value becomes smaller.

4. A method according to claim 1, further comprising the step of designating one of the character candidates based on the result of determining the rank of the character candidates.

5. A method according to claim 4, further comprising the steps of checking, when one of the character candidates other than the character candidate having a first rank is designated, whether there exists a cluster of one of the character candidates which is different from the designated one character candidate and has a distance value smaller than the distance value of the designated one character candidate, and if exists, newly registering in the weighting dictionary the segment of the ink data segmented at one of the position candidates being the most remote from the leading position of the ink data as the weighting cluster together with the character candidate having a higher rank than the designated character candidate as the object character.

6. A method according to claim 5, further comprising the step of erasing an old weighting cluster, which has been previously registered in the weighting dictionary.

7. A handwritten character recognition device comprising:

input means for inputting by a user characters in a handwriting manner;

storage means for storing ink data of the inputted handwritten characters;

a recognition dictionary in which a plurality of characters and corresponding character clusters are registered;

a weighing dictionary in which weighing clusters and corresponding object characters are registered;

means for determining all segmentation position candidates, which satisfy a predetermined condition, for segmentation of the ink data stored in the storage means along a direction corresponding to a direction in which the ink data is inputted from a leading position of the ink data;

means for obtaining character candidates corresponding to segments of the ink data segmented at the segmentation position candidates, respectively, based upon a distance value between each of the segments of the ink data and the character cluster of a corresponding one of the character candidates by referring to the recognition dictionary;

means for applying a weighing quantity to the distance value between any one of the segments of the ink data and the character cluster of the corresponding one of the character candidates, if a distance between one of the weighing clusters included in the weighing dictionary and said one of the segments of the ink data is equal to or less than a predetermined threshold value and if the object character of said one weighing cluster coincides with said corresponding one of the character candidates; and

means for ranking the character candidates based on the distance values, or the weighted distance values, if any, thereof.

8. A handwritten character recognition apparatus according to claim 7, further comprising means for checking, when one of the character candidates ranked with a first order is different from one of the character candidates which is intended by the user, whether there exists a cluster of one of the character candidates which is different from the character candidate intended by the user and has a distance value smaller than the distance value of the user'"'"'s intended character candidate, and if exists, newly registering in the weighting dictionary the segment of the ink data segmented at one of the position candidates being the most remote from the leading position of the ink data as the weighting cluster together with the character candidate ranked higher than the user'"'"'s intended character candidate as the object character.

9. A handwritten character recognition apparatus according to claim 8, further comprising means for erasing an old weighting cluster, which has been previously registered in the weighting dictionary.

10. A method of recognizing handwritten characters, comprising the steps of:

preparing a recognition dictionary in which a plurality of characters and character feature vectors corresponding to said plurality of characters, respectively, are registered and a weighing dictionary in which a plurality of weighing clusters and object characters corresponding to said weighing clusters, respectively, are registered;

inputting handwritten character ink data via input means into processing means;

separating said inputted handwritten character ink data at segmentation position candidates to obtain a plurality of ink data segments, said separation being started with a leading position of said inputted handwritten character ink data;

calculating a feature vector for each of the ink data segments;

determining a distance value between the feature vector obtained for each of said ink data segments and a selected one of the character feature vectors registered in said recognition dictionary, said distance being limited to a predetermined range;

selecting a character candidate corresponding to each of said ink data segments based on said distance value;

searching the weighing dictionary and applying a weighing quantity to said distance value determined for one of the character feature vectors corresponding to said selected character candidate, (a) if a distance between at least one of the weighing clusters registered in said weighing dictionary and the feature vector of one of said ink data segments is equal to or less than a predetermined threshold value, and (b) if the object character corresponding to said at least one weighing cluster coincides with said selected character candidate; and

determining a rank of the character candidates in accordance with the weighted distance value thereof, whereby said ink data segment is recognized based on said rank.