Handwriting recognition using neural networks

• US 20070009151A1
• Filed: 06/23/2005
• Published: 01/11/2007
• Est. Priority Date: 06/23/2005
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Manu, Mitica, Pittman, James
• Granted Patent: US 7,428,516 B2
• Time in Patent Office: Days
• US Class Current: 382/159
• CPC Class Codes: G06V 30/244 using graphical properties, e.g. alphabet type or font
• CPC Class Codes: G06V 30/373 using a special pattern or subpattern alphabet

Abstract

New neural networks for handwriting recognition may be build from existing neural networks. An existing neural network pre-trained for a starting language is chosen based on a desired target language. The neural network is modified so that it may be used to recognize characters of the target language, and the modified neural network is used in a handwriting recognizer for the target language. Modification includes copying one or more of the primary outputs of the existing neural network. An appropriate starting language may be chosen based on the desired target language. In addition, a “super network” may be provided that is a relatively large neural network configured to recognize characters from a number of different languages. One may customize a handwriting recognizer using such a super network by programming a mask to block outputs from the super network that are not necessary for the language desired to be recognized.

Claims

1. A method for providing handwriting recognition functionality, comprising:

providing a neural network having a plurality of primary outputs each corresponding to a different character of a first language;

choosing a first primary output from the plurality of primary outputs; and

adding a copy of the first primary output to the neural network.

2. The method of claim 1, wherein the neural network has a plurality of primary inputs, and wherein the copy of the first primary output responds to the primary inputs in an identical manner as the first primary output.

3. The method of claim 1, further including, after the step of adding, training the neural network such that the copy of the first primary output responds to the primary inputs in a manner different from the first primary output.

4. The method of claim 1, further including selecting the neural network from a plurality of neural networks, each of the plurality of neural networks being associated with a different language.

5. The method of claim 1, further including:

choosing the first language; and

choosing a second language different from the first language,

wherein the step of choosing the first primary output includes choosing the first primary output depending upon the second language.

6. The method of claim 5, wherein the step of choosing the first primary output includes:

comparing a character of the first language with a character of the second language; and

if the character of the first language is visually similar, but not identical, to the character of the second language, then copying the first primary output.

7. The method of claim 5, further including providing a handwriting recognizer that includes the neural network, including the copy of the first primary output, and a language model of the second language.

8. The method of claim 1, further including providing a handwriting recognizer utilizing the neural network including the copy of the first primary output.

9. At least one computer-readable medium storing computer-executable instructions for providing handwriting recognition functionality, wherein the computer-executable instructions are for performing steps comprising:

choosing a first network from a plurality of networks, wherein the first network is associated with a first language, and wherein the first network has a plurality of primary outputs each representing a different character of the first language; and

choosing a first primary output from the plurality of primary outputs of the first network, the first primary output being associated with one of the characters of the first language.

10. The at least one computer-readable medium of claim 9, wherein the computer-executable instructions are further for performing steps including:

receiving a first user input indicating a second language; and

choosing the first language based on the second language.

11. The method of claim 10, wherein the step of choosing the first primary output includes:

comparing a character of the first language with a character of the second language; and

if the character of the first language is visually similar, but not identical, to the character of the second language, then copying the first primary output.

12. The at least one computer-readable medium of claim 9, wherein the computer-executable instructions are further for adding a copy of the first primary output to the first network.

13. The at least one computer-readable medium of claim 9, wherein the first network has a plurality of primary inputs, and wherein the copy of the first primary output responds to the primary inputs in a manner identical to the first primary output.

14. The at least one computer-readable medium of claim 9, wherein after the first primary output is chosen, the first network includes a primary output that is a copy of the first primary output, wherein the first network is a neural network, and wherein the computer-executable instructions are further for performing steps including:

receiving a second user input comprising handwritten ink, the handwritten ink including a representation of a character of the second language that is not part of the first language; and

in response to the second user input, training the first network such that a response of the copy of the primary output is modified.

15. The at least one computer-readable medium of claim 9, the at least one computer-readable medium further storing the plurality of networks.

16. The at least one computer-readable medium of claim 9, wherein the first network is a neural network, and the networks of the plurality of networks are each a neural network.

18. The handwriting recognizer of claim 16, wherein the network, the post-processing unit, and the masking unit comprise software.

19. The handwriting recognizer of claim 16, wherein the first subset includes those of the primary outputs that represent characters of the first language but not the second language, and wherein the second subset includes those of the primary outputs that represent characters of the second language but not the first language.

20. The handwriting recognizer of claim 16, wherein the network is a neural network.

17. A handwriting recognizer, comprising:

a network having a first plurality of primary outputs each representing a different character of a first language and a second plurality of primary outputs each representing a different character of a second language;

a post-processing unit configured to interpret signals on the primary outputs; and

a masking unit configured to selectively block one of a first subset or a second subset of the primary outputs from being interpreted by the post-processing unit.