Recognition method and system

Abstract

Techniques for recognizing discrete multi-component symbolic input from a user can be applied to, for example, handwriting or speech. The techniques can include providing a database of model input sequences, where each model input sequence corresponds to a symbol to be recognized. Input functions, for example, discrete strokes, are obtained from a user and segmented into a sequence of discrete components. Hypothesis symbol sequences are obtained by comparing the discrete components to a database of symbols to be recognized and updating hypothesis symbol sequences based on the results of the comparison and hypothesis symbol sequence history from input previously acquired in time.

1 - 22 . (canceled) 
     
     
         23 . A method ( 70 ) of creating a handwriting recognition database comprising the steps of:
 a) acquiring ( 72 ) spatiotemporal_training input from a user corresponding to an exemplar character, wherein the spatiotemporal input is provided in the form of discrete input strokes;   b) normalizing ( 74 ) the discrete input strokes into a sequence of normalized representations; and   c) storing ( 76 ) the normalized representations into the database.   
     
     
         24 . The method of  claim 23 , wherein the step of acquiring training spatiotemporal input comprises separating the spatiotemporal input into the discrete input strokes. 
     
     
         25 . The method of  claim 23 , wherein the step of normalizing the discrete input strokes comprises forming a non-uniform rational b-spline corresponding to the input stroke. 
     
     
         26 . The method of  claim 23 , wherein the non-uniform rational b-spline is normalized to scale it to fit between 0 and 1 in all parameters and coordinates. 
     
     
         27 . The method of  claim 23 , further comprising storing more than one normalized representation for a given exemplar character within the database. 
     
     
         28 . A method of recognizing handwritten input, comprising:
 a) providing a trellis definition ( 50 ) having a plurality of nodes corresponding to a plurality of written_characters to be recognized, each character defined by at least one discrete stroke element;   b) acquiring ( 34 ) written_spatiotemporal input from a user in the form of discrete input strokes; and   c) defining and updating a plurality of node scores ( 60 ) for each discrete input stroke, wherein the node scores are advanced non-uniformly in time through the trellis.   
     
     
         29 . The method of  claim 28  wherein the step of updating a plurality of node scores comprises updating each node score based on a node score obtained previously in time from a number of strokes corresponding to a node hypothesis character. 
     
     
         30 - 37 . (canceled)