Adapting a neural network for individual style

• US 7,702,145 B2
• Filed: 06/28/2006
• Issued: 04/20/2010
• Est. Priority Date: 06/28/2006
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Goval, Manish, Revow, Michael
• Primary Examiner(s): Desire; Gregory M
• Application Number:
  US11/477,332
• Publication Number:
  US 20080002886A1
• Time in Patent Office: 1,392 Days
• Field of Search: 382/155, 382/156, 382/159, 382/161, 382/179, 382/186, 382/187, 382/189
• US Class Current: 382/156
• CPC Class Codes: G06F 18/28 Determining representative reference patterns, e.g. by averaging or distorting; Generating dictionaries
• CPC Class Codes: G06V 30/1914 Determining representative reference patterns, e.g. averaging or distorting patterns; Generating dictionaries, e.g. user dictionaries
• CPC Class Codes: G06V 30/36 Matching; Classification

Abstract

Various technologies and techniques are disclosed for improving handwriting recognition using a neural network by allowing a user to provide samples. A recognition operation is performed on the user'"'"'s handwritten input, and the user is not satisfied with the recognition result. The user selects an option to train the neural network on one or more characters to improve the recognition results. The user is prompted to specify samples for the certain character, word, or phrase, and the neural network is adjusted for the certain character, word, or phrase. Handwritten input is later received from the user. A recognition operation is performed on the handwritten input using the neural network that was adjusted for the certain character or characters.

Claims

1. A method for adjusting a writer-independent neural network based on user-supplied samples comprising the steps of:

receiving handwritten input from a user;

performing a recognition operation on the handwritten input;

displaying a recognition result to the user;

receiving a selection from the user to train a neural network on at least one certain character because the user was not satisfied with the recognition result;

prompting the user to supply one or more samples for the at least one certain character;

removing an effect of any writer-independent character samples on the neural network for the at least one certain character leaving only the one or more samples of the at least one certain character supplied by the user; and

adjusting the neural network for the at least one certain character.

2. The method of claim 1, further comprising:

using the adjusted neural network for performing a later recognition operation.

3. The method of claim 2, wherein the later recognition operation involves the user inputting the at least one certain character, and wherein a correct result is returned by a recognizer based on a fact that the user previously trained the neural network on the at least one certain character.

4. The method of claim 2, further comprising:

displaying a result of the later recognition operation to the user on a display device.

5. The method of claim 1, wherein the neural network is adjusted for the at least one certain character by adjusting at least one decision boundary for the at least one certain character based on the one or more samples supplied by the user.

6. The method of claim 1, wherein at least one decision boundary for the at least one certain character is adjusted after the one or more samples supplied by the user are added to the neural network.

7. The method of claim 1, wherein the at least one certain character is selected from a group consisting of a single character, multiple characters, a word, and a phrase.

8. The method of claim 1, wherein at least one decision boundary for the at least one certain character is adjusted after adjusting at least one parameter of the neural network in response to the one or more samples supplied by the user.

9. A computer-readable medium having computer-executable instructions for causing a computer to perform the steps recited in claim 1.

10. The method of claim 1, further comprising:

selecting a particular character;

extracting well recognized instances of the particular character from a database;

forming a plurality of clusters based on the extracted well recognized instances of the particular character; and

saving a small number of samples from each of the plurality of clusters in proportion to a frequency of the respective clusters in a general population.

11. A computer-readable medium having computer-executable instructions for causing a computer to perform steps comprising:

receive a selection from a user to train a writer-independent neural network on a certain character for which the user is not satisfied with a recognition result from a recognizer;

prompt the user to specify one or more samples for the certain character;

remove an effect of any writer-independent character samples on the neural network for the certain character leaving only the one or more samples of the certain character supplied by the user;

adjust the neural network based upon the one or more samples supplied by the user for the certain character;

receive handwritten input from the user; and

perform a recognition operation on the handwritten input using the adjusted neural network.

12. The computer-readable medium of claim 11, further having computer-executable instructions for causing a computer to perform the step comprising:

display a result of the recognition operation on a display device.

13. The computer-readable medium of claim 11, further having computer-executable instructions for causing a computer to perform the step comprising:

leave decision boundaries of the neural network intact for one or more other characters for which the user did not supply one or more samples.

14. A method for improving handwriting recognition results for user-supplied samples comprising the steps of:

removing an effect of nonuser-supplied character samples on a neural network for a certain character leaving only one or more samples provided by a user for the certain character;

adjusting the neural network for the certain character with the one or more samples provided by the user for the certain character;

receiving handwritten input from the user; and

performing a recognition operation on the handwritten input using the neural network that was adjusted for the certain character.

15. The method of claim 14, further comprising:

displaying a result of the recognition operation to the user on a display device.

16. The method of claim 14, wherein the neural network is adjusted for the certain character by adjusting at least one decision boundary associated with the certain character in the neural network.

17. The method of claim 14, wherein the adjusting is performed by adjusting at least one parameter associated with the neural network to take into account the one or more samples provided by the user.

18. A computer-readable medium having computer-executable instructions for causing a computer to perform the steps recited in claim 14.

19. The method of claim 14, further comprising:

supplementing a set of nonuser-supplied character samples, which were used to adapt the neural network for characters other than the certain character, with other nonuser-supplied samples from a pool of characters.