Automatic handwriting recognition using both static and dynamic parameters

Claim 1

A method for processing input signals representative of a hand written character, the input signals being generated by a handwriting transducer in response to a writer and being ordered as points each having an x-axis coordinate value and a y-axis coordinate value, comprising the steps of: coupling a handwriting processor to an output of the handwriting transducer for receiving the input signals; operating the handwriting processor to execute the steps of, (a) sorting points received from the input signals in terms of x-axis values; (b) sampling along the x-axis at intervals of δx to provide a sequence of the form {x i }, where x i =x i-1 +δx; (c) defining, for each sample point x i , a slice of width lw centered on that sample and associating all y-axis values within the width of the slice with that sample, thereby yielding a vector of the form y(x i ); (d) quantizing a range of y-axis values to n y equispaced levels, l k , and assigning, for each x i , all associated y-axis values to one level in accordance with the expression: yεl.sub.k l.sub.k-1 Δl<y<l.sub.k Δl, where Δl=(ymax-ymin)/n y ; (e) constructing a bitmap of the slice having a grid size n y , the step of constructing including a step of determining, for each x i , a first feature vector fx(x i ) of length n y , such that the kth element is one if at least one y-axis value was assigned to l k , and zero otherwise; (f) constructing, for each x i , a second feature vector of length 1, f cg (x i ) such that: ##EQU11## where the summation is over all y-axis values within the slice associated with x i ; (g) determining a subset of points of {x i } such that the points are spaced ax apart, the subset of points being defined as {X i }; (h) constructing, at each location X i , an N dimensional spliced vector by concatenating together H x feature vectors f x preceding a current point X i , a feature vector f(X i ) at the current point, and H x feature vectors succeeding X i to yield a spliced feature vector, F(x i ), of length: N=n.sub.y (2H.sub.x +1); (i) repeating Step (h) for f cg (x i ) to obtain F cg (x i ); (j) repeating steps (a) through (e) and (g) through (h) for the y-axis dimension to obtain a feature vector, F y (Y i ); and further operating the handwriting processor to determine a most probable identification of the writer's handwriting in accordance with the obtained feature vectors.

Claim 2

A method for processing input signals representative of a hand written character, the input signals being generated by a handwriting transducer in response to a writer and being ordered as points each having an x-axis coordinate value and a y-axis coordinate value, comprising the steps of: coupling a handwriting processor to an output of the handwriting transducer for receiving the input signals; operating the handwriting processor to execute the steps of, (a) sorting points received from the input signals in terms of x-axis values; (b) sampling along the x-axis at intervals of δx to provide a sequence of the form {x i }, where x i =x i-1 +δx; (c) defining, for each sample point x i , a slice of width lw centered on that sample and associating all y-axis values within the width of the slice with that sample, thereby yielding a vector of the form y(x i ); (d) quantizing a range of y-axis values to n y equispaced levels, l k , and assigning, for each x i , all associated y-axis values to one level in accordance with the expression: yεl.sub.k l.sub.k-1 Δl<y<l.sub.k Δl, where Δl=(ymax-ymin)/n y ; (e) constructing a bitmap of the slice having a grid size n y , the step of constructing including a step of determining, for each x i , a first feature vector fx(x i ) of length n y , such that the kth element is one if at least one y-axis value was assigned to l k , and zero otherwise; (f) constructing, for each x i , a second feature vector of length 1, f cg (x i ) such that: ##EQU11## where the summation is over all y-axis values within the slice associated with x i ; (g) determining a subset of points of {x i } such that the points are spaced ax apart, the subset of points being defined as {X i }; (h) constructing, at each location X i , an N dimensional spliced vector by concatenating together H x feature vectors f x preceding a current point X i , a feature vector f(X i ) at the current point, and H x feature vectors succeeding X i to yield a spliced feature vector, F(x i ), of length: N=n.sub.y (2H.sub.x +1); (i) repeating Step (h) for f cg (x i ) to obtain F cg (x i ); (j) repeating steps (a) through (e) and (g) through (h) for the y-axis dimension to obtain a feature vector, F y (Y i ); and further operating the handwriting processor to determine a most probable identification of the writer's handwriting in accordance with the obtained feature vectors.