Method and system for extracting features from handwritten text

• US 5,757,960 A
• Filed: 02/28/1997
• Issued: 05/26/1998
• Est. Priority Date: 09/30/1994
• Status: Expired due to Term

• Current Assignee: Google Technology Holdings LLC (Alphabet Inc.)
• Original Assignee: Michael Chase Murdock, Nicholas Mikulas Labun, Shay-Ping Thomas Wang
• Inventors: Wang, Shay-Ping Thomas, Murdock, Michael Chase, Labun, Nicholas Mikulas
• Primary Examiner(s): Johns, Andrew W.
• Assistant Examiner(s): Davis, Monica S.
• Application Number:
  US08/808,707
• Time in Patent Office: 452 Days
• Field of Search: 382/122, 382/186, 382/187, 382/190, 382/191, 382/195
• US Class Current: 382/187
• CPC Class Codes: G06V 30/196 using sequential comparisons of the image signals with a plurality of references
• CPC Class Codes: G06V 30/333 Preprocessing; Feature extraction
• CPC Class Codes: G06V 30/36 Matching; Classification

Abstract

A handwriting recognition system achieves a higher recognition rate by using a feature extraction method which computes features based on multiple data frames. A plurality of data frames is generated from handwritten text received by the system. Each data frame includes samples taken from the handwritten text. Individual-frame features are extracted from individual data frames, and in turn, multi-frame features are extracted from individual-frame features which correspond to different data frames.

Claims

1. In a system for recognizing a plurality of characters from a sample of handwritten text, the system utilizing a classifier which responds to a plurality of features extracted from the sample of handwritten text, a method for extracting the plurality of features, the method comprising the steps of:

(a) receiving the sample of handwritten text;

(b) sampling the handwritten text, over time, to form a sequence of sample datum;

(c) partitioning the sequence of sample datum into a temporal sequence of data frames, each of the temporal sequence of data frames including at least two of the sequence of sample datum;

(d) extracting a plurality of individual-frame feature from the temporal sequence of data frames, each of the plurality of individual-frame features having a magnitude and corresponding to one of the temporal sequence of data frames, wherein at least one of the individual-frame features includes a plurality of coefficients of a first order polynomial which is fitted to a curvilinear velocity profile;

space="preserve" listing-type="equation">v.sub.k =a.sub.0 +a.sub.1 v.sub.(k-1) +a.sub.2 v.sub.(k-2) +a.sub.3 v.sub.(k-3)

wherein v.sub.(k) represents the curvilinear velocity of a kth sample datum, v.sub.(k-1) represents the curvilinear velocity of a (k-1)th sample datum, v.sub.(k-2) represents the curvilinear velocity of a (k-2)th sample datum, v.sub.(k-3) represents the curvilinear velocity of a (k-3)th sample datum, k is an integer index, and a₀, a₁, a₂, and a₃ represent the coefficients of the first order polynomial; and

(e) extracting a multi-frame feature, corresponding to a specific data frame of the temporal sequence of data frames, from one of;

at least two of the plurality of individual-frame features,

at least two of the temporal sequence of data frames, and

at least one of the plurality of individual-frame features and at least one of the temporal sequence of data frames.

2. The method of claim 1 wherein the plurality of individual-frame features includes a feature selected from the group consisting of angular velocity, curvilinear velocity, stroke angle, stroke curvature, data frame index, and Fourier coefficients.

3. The method of claim 1 wherein step (d) includes deriving the plurality of individual-frame features from coordinates of the sample of handwritten text, over time, as the sample is written.

4. The method of claim 1 wherein the step of extracting a multi-frame feature includes determining whether a difference between the magnitude of an individual-frame feature extracted from the specific data frame and the magnitude of an individual-frame feature extracted from a data frame adjacent to the specific data frame is less than a threshold value:

(i) if so, setting the multi-frame feature to 1; and

(ii) if not, setting the multi-frame feature to 0.

5. The method of claim 1 wherein the step of extracting a multi-frame feature includes setting the multi-frame feature to n, wherein n is a positive integer which indicates that the magnitude of each of the plurality of individual-frame features corresponding to n data frames precedingly adjacent to the specific data frame has persisted within a range.

6. In a system for recognizing a plurality of characters from a sample of handwritten text, the system utilizing a classifier which responds to a plurality of features extracted from the sample of handwritten text, a method for extracting the plurality of features, the method comprising the steps of:

(a) receiving the sample of handwritten text;

(b) sampling the handwritten text, over time, to form a sequences of sample datum,

(c) partitioning the sequence of sample datum into a temporal sequence of data frames, each of the temporal sequence of data frames including at least two of the sequence of sample datum;

(d) extracting a plurality of individual-frame feature from the temporal sequence of data frames, each of the plurality of individual-frame features having a magnitude and corresponding to one of the temporal sequence of data frames, wherein at least one of the individual-frame features includes a plurality of coefficients of a first order polynomial which is fitted to a angular velocity profile;

space="preserve" listing-type="equation">v.sub.k =b.sub.0 +b.sub.1 v.sub.(k-1) +b.sub.2 v.sub.(k-2) +b.sub.3 v.sub.(k-3)

wherein v.sub.(k) represents the angular velocity of a kth sample datum, v.sub.(k-1) represents the angular velocity of a (k-1)th sample datum, v.sub.(k-2) represents the angular velocity of a (k-2)th sample datum, v.sub.(k-3) represents the angular velocity of a (k-3)th sample datum, k is an integer index, and b₀, b₁, b₂, and b₃ represent the coefficients of the first order polynomial; and

(e) extracting a multi-frame feature, corresponding to a specific data frame of the temporal sequence of data frames, from one of;

at least two of the plurality of individual-frame features,

at least two of the temporal sequence of data frames, and

at least one of the plurality of individual-frame features and at least one of the temporal sequence of data frames.

7. The method of claim 6 wherein the plurality of individual-frame features includes a feature selected from the group consisting of angular velocity, curvilinear velocity, stroke angle, stroke curvature, data frame index, and Fourier coefficients.

8. The method of claim 6 wherein step (d) includes deriving the plurality of individual-frame features from coordinates of the sample of handwritten text, over time, as the sample is written.

9. The method of claim 6, wherein step (e) includes the sub-step of:

generating the multi-frame feature based on a comparison of magnitudes of at least two of the individual-frame features.

10. The method of claim 6 wherein step (e) includes the sub-step of:

setting the multi-frame feature to n, wherein n is a positive integer which indicates that the magnitude of each of the plurality of individual-frame features corresponding to n data frames precedingly adjacent to the specific data frame has persisted within a range.

11. In a system for recognizing handwritten text, a sub-system for extracting a plurality of features, the sub-system comprising:

a frame extractor partitioning a sequence of sample datum derived from the handwritten text into a temporal sequence of data frames, each of the temporal sequence of data frames including at least two of the sequence of sample datum; and

a feature extractor for extracting a plurality of individual-frame features and a multi-frame feature from the temporal sequence of data frames, wherein at least one of the individual-frame features includes a plurality of coefficients of a first order polynomial which is fitted to a velocity profile

space="preserve" listing-type="equation">v.sub.k =b.sub.0 +b.sub.1 v.sub.(k-1) +b.sub.2 v.sub.(k-2) +b.sub.3 v.sub.(k-3)

wherein v.sub.(k) represents the velocity of a kth sample datum, v.sub.(k-1) represents the velocity of a (k-1)th sample datum, v.sub.(k-2) represents the velocity of a (k-2)th sample datum, v.sub.(k-3) represents the velocity of a (k-3)th sample datum, k is an integer index, and b₀, b₁, b₂, and b₃ represent the coefficients of the first order polynomial.

12. The sub-system of claim 11 wherein the velocity profile is an angular velocity profile.

13. The sub-system of claim 11 wherein the velocity profile is an curvilinear velocity profile.

14. The sub-system of claim 11 wherein the plurality of individual-frame features includes a feature selected from the group consisting of angular velocity, curvilinear velocity, stroke angle, stroke curvature, data frame index, and Fourier coefficients.

15. The sub-system of claim 11 wherein the plurality of individual-frame features are extracted from the sample of handwritten text as the sample is written.

16. The sub-system of claim 11 wherein the feature extractor generates a multi-frame feature based on a comparison of at least two of the individual-frame features.

17. The sub-system of claim 11 wherein the feature extractor sets a multi-frame feature to n, wherein n is a positive integer which indicates that a sequence of n individual-frame features has persisted within a range.