Handwriting recognition using a comparative neural network

• US 20070005537A1
• Filed: 06/10/2005
• Published: 01/04/2007
• Est. Priority Date: 06/02/2005
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Zhang, Qi, Abdulkader, Ahmad, Drakopoulos, Ioannis
• Granted Patent: US 7,496,547 B2
• Time in Patent Office: Days
• US Class Current: 706/20
• CPC Class Codes: G06F 18/28 Determining representative reference patterns, e.g. by averaging or distorting; Generating dictionaries
• CPC Class Codes: G06F 18/41 Interactive pattern learning with a human teacher
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: G06V 10/20 Image preprocessing
• CPC Class Codes: G06V 30/287 of Kanji, Hiragana or Katakana characters
• CPC Class Codes: G06V 30/373 using a special pattern or subpattern alphabet
• CPC Class Codes: G06V 30/387 using human interaction, e.g. selection of the best displayed recognition candidate

Abstract

Handwriting recognition techniques employing a personalized handwriting recognition engine. The recognition techniques use examples of an individual'"'"'s previous writing style to help recognize new pen input from that individual. The techniques also employ a shape trainer to select samples of an individual'"'"'s handwriting that accurately represent the individual'"'"'s writing style, for use as prototypes to recognize subsequent handwriting from the individual. The techniques also alternately or additionally employ an intelligent combiner to combine the recognition results from the personalized recognition engine and the conventional recognition engine (or engines). The combiner may use a comparative neural network to combine the recognition results from multiple recognition engines. The combiner alternately may use a rule-based system based on prior knowledge of different recognition engines.

Claims

1. A method of comparing inputs, comprising:

receiving a first input;

receiving a second input;

ordering the first input with respect to the second input using a neural network.

2. The method recited in claim 1, wherein each of the first input and the second input is a list of one or more alternate classifications output by a classifier.

3. The method recited in claim 2, wherein the classifier includes a handwriting recognition engine, and the one or more alternate classifications are characters recognized from handwriting.

4. The method recited in claim 3, further comprising outputting a character that is a highest-ranked choice from among the lists.

5. The method recited in claim 2, further comprising outputting an ordering of the one or more alternates from both lists.

6. The method recited in claim 1, wherein the neural network employs a mergesort process to compare the first input with respect to the second input.

7. A method of training a neural network to compare a first input with a second input, comprising:

(a) selecting an input x;

(b) identifying a correct classification a(x) for the input x;

(c) operating each of a plurality of base-classifiers to obtain an ordered list of alternative classifications of x from each base-classifier;

(d) merge the alternative lists together such that common classifications are coalesced;

(e) if the correct classification σ

(x) does not appear in the merged list, discarding the merged list;

(f) if the correct classification σ

(x) does appear in the merged list, then extracting context features C(x) from x, and

combining σ

(x) with all erroneous classifications σ

(x) in a combined list so as to create comparative training samples of the form (C(x), D(σ

(x)), D(λ

(x)),

     0) or (C(x), D(λ

(x)), D(σ

(x)),

     1) where D( ) is a function the maps classifications to their description features;

(g) executing steps (a)-(f) through one or more base-classifier training sets;

(h) collecting the generated comparative training samples into an single training set;

(i) shuffling the set; and

(j) training a comparative neural network with the set.

8. A classification tool, comprising:

a first classifier that produces a first classifier output;

a second classifier that produces a second classifier output; and

a comparative neural network that compares the classifier output with the second classifier output.

9. The classification tool recited in claim 8, wherein each of the first classifier output and the second classifier output is a list of one or more alternate classifications.

10. The method recited in claim 9, wherein

the first classifier and the second classifier each includes a handwriting recognition engine, and

the one or more alternate classifications are characters recognized from handwriting

11. The classification tool recited in claim 8, wherein the neural network is configured to employ a mergesort process to compare the first classifier output with respect to the second classifier output.