EP0280064A1 — Pattern matching system

Claim 1

Independent Claim 192 words

A pattern matching system comprising: distance calculating means for obtaining the distance between feature parameters of an input pattern and a reference pattern which are time series of feature parameter vectors at each time point; integrating means for obtaining an integration value which is an accumulated distance of a first distance at a first time point and a second distance at each of a plurality of second time points before said first time point, said first and second distances being obtained by said distance calculating means; a path cost memory for storing a path cost attached, at each time point of said reference pattern, to each of a plurality of predetermined paths from said second time points to said first time point; path cost reading means for reading from said path cost memory a path cost in response to the time point of said reference pattern; weighting means for weighting the path cost thereby read out to said integration value; and optimum path selecting means for selecting a path from one of said second time points to said first time point having the minimum weighted integration value to provide the integration value.

Claim 2

Dependent Claim 192 words

A pattern matching system comprising: distance calculating means for obtaining the distance between feature parameters of an input pattern and a reference pattern which are time series of feature parameter vectors at each time point; integrating means for obtaining an integration value which is an accumulated distance of a first distance at a first time point and a second distance at each of a plurality of second time points before said first time point, said first and second distances being obtained by said distance calculating means; a path cost memory for storing a path cost attached, at each time point of said reference pattern, to each of a plurality of predetermined paths from said second time points to said first time point; path cost reading means for reading from said path cost memory a path cost in response to the time point of said reference pattern; weighting means for weighting the path cost thereby read out to said integration value; and optimum path selecting means for selecting a path from one of said second time points to said first time point having the minimum weighted integration value to provide the integration value.

Claim 3

Dependent Claim 47 words

A pattern matching system according to claim l, further comprising a recognition means for obtaining the integration value of the distance between the input pattern and a plurality of prepared reference patterns, and recognizing a reference pattern having the minimum distance integration value as the input pattern.

Claim 4

Dependent Claim 44 words

A pattern matching system according to claim l or 2, wherein said path cost is determined on the basis of a distribution of a plurality of training patterns in the same category of said reference pattern at each time point of said reference pattern.

Claim 5

Dependent Claim 31 words

A pattern matching system according to claim 3, wherein said distribution is the deviation distribution of eachof said plurality of training patterns from a representative pattern selected from said training patterns.

Claim 6

Dependent Claim 22 words

A pattern matching system according to claim 4, wherein said representative pattern is an average pattern of said plurality of training patterns.

Claim 7

Dependent Claim 55 words

A pattern matching system according to claim 4 or 5, wherein said path cost is determined on the basis of the occurrence rate distribution of a matching path slope at each time point of said representative pattern obtained by time-normalization of each of said training patterns relative to said representative pattern through dynamic programming processing.

Claim 8

Dependent Claim 87 words

A pattern matching system according to claim 6, wherein a path cost P r (j) of a path r at a time point j is determined on the basis of an occurrence rate distribution W of said path r by P r (j)= -α · logW where W = W r (j)/W(j) , W r (j) represents the occurrence rate of said path r at said time point j, W(j) represents the sum of the occurrence rates of all paths, and α represents a positive weighing coefficient.

Claim 9

Dependent Claim 159 words

A pattern matching system comprising: an input pattern memory for memorizing a time series A = {a(i), i-l, ..., I} of feature vectors of an input pattern; a reference pattern memory for memorizing a time series B = {b(i), j-l, ..., J} of feature vectors of a reference pattern; distance calculating means for obtaining a distance d(i, j) at a time point (i, j) between two patterns read out from said two memories; a path cost memory for memorizing a path cost which is a predetermined value attached at said time point j to each of a plurality of paths from plurality of time points before said time point (i, j) to said time point (i, j); and weighting means for obtaining a weighted value by weighting a path cost read out from said path cost memory at said time point j to an integration distance value obtained at a time point on one of said plurality of paths.

Claim 10

Dependent Claim 159 words

A pattern matching system comprising: an input pattern memory for memorizing a time series A = {a(i), i-l, ..., I} of feature vectors of an input pattern; a reference pattern memory for memorizing a time series B = {b(i), j-l, ..., J} of feature vectors of a reference pattern; distance calculating means for obtaining a distance d(i, j) at a time point (i, j) between two patterns read out from said two memories; a path cost memory for memorizing a path cost which is a predetermined value attached at said time point j to each of a plurality of paths from plurality of time points before said time point (i, j) to said time point (i, j); and weighting means for obtaining a weighted value by weighting a path cost read out from said path cost memory at said time point j to an integration distance value obtained at a time point on one of said plurality of paths.

Claim 11

Dependent Claim 55 words

A pattern matching system according to claim 8, wherein, under initial conditions that g(O, O) = 0, g(i, O) = ∞ (i > O), said weighting means obtains said weighted value at said time point (i, j) by where P₁(j), P₂(j), P₃(j) are path costs attached to three paths to said time point (i, j).

Claim 12

Dependent Claim 65 words

l0. A pattern matching system in which the distance between time points of an input pattern and a reference pattern is obtained, time points at which each of the accumulated distances exhibits the minimum are connected to form a matching path, said system limits the advancement slope angle of said matching path to a predetermined angle region determined at each time of said reference pattern.

Claim 13

Dependent Claim 65 words

l0. A pattern matching system in which the distance between time points of an input pattern and a reference pattern is obtained, time points at which each of the accumulated distances exhibits the minimum are connected to form a matching path, said system limits the advancement slope angle of said matching path to a predetermined angle region determined at each time of said reference pattern.

Pattern matching system

Abstract

Inventors

Bibliographic Details

Claims Analysis

References Cited (Backward Citations)

USPTO Annuity & Maintenance Windows

Assigned Categories & Tags

Assign Category Tag