Patent EP0374604A2 successfully refreshed from Google Patents.
EP0374604A2
Expired / Lapsed
EP
⏱️ 0.0 years remaining
Pattern recognition system and method
Assignee: Toshiba Corp
•
Filing: 1989-12-07
•
Granted: —
•
Expires: 2009-12-07
Abstract
Abstract An inner product computing unit (12) computes inner products of an input pattern whose category is unknown, and orthogonalized dictionary sets of a plurality of reference patterns whose categories are known. A nonlinear converting unit (14) nonlinearly converts the inner products in accordance with a positive-negative symmetrical nonlinear function. A neural network unit (17) or a statistical discriminant function computing unit (19) performs predetermined computations of the nonlinearly converted values on the basis of preset coefficients in units of categories using a neural network or a statistical discriminant function. A determining section (18) compares values calculated in units of categories using the present coefficients with each other to discriminate a category to which the input pattern belongs.
Inventors
- 👤 Tsuneo Intellectual Property Division Nitta
Bibliographic Details
Application Number
—
Jurisdiction
EP
Data Source
google_patents
Last Synchronized
2026-11-09 14:54
CPC Classifications
G; G10; G10L; G10L15/00; G10L15/08; G10L15/16; G06; G06F; G06F18/00; G06F18/20; G06F18/22; G06N; G06N3/00; G06N3/02; G06N3/04; G06N3/045; G06N3/0499; G06N3/08; G06N3/09
Claims Analysis
Total Claims: 14
Claim 1
Independent Claim
152 words
A pattern recognition system comprising: orthogonalized dictionary means for storing a plurality of reference patterns whose categories are known; inner product computing means for computing inner products of an input pattern whose category is unknown and the plurality of reference patterns which are stored in said orthogonalized dictionary means and whose categories are known; converting means for nonlinearly converting the inner products, which are computed by said inner product computing means, in accordance with a positive-negative symmetrical nonlinear function; evaluation value computing means for computing evaluation values on the basis of the nonlinearly converted values by said converting means and coefficients which are preset in units of categories; comparing means for comparing the evaluation values obtained by said evaluation value computing means in units of categories with each other; and discriminating means for discriminating a category to which the input pattern belongs on the basis of comparison results of said comparing means.
Claim 2
Dependent Claim
152 words
A pattern recognition system comprising: orthogonalized dictionary means for storing a plurality of reference patterns whose categories are known; inner product computing means for computing inner products of an input pattern whose category is unknown and the plurality of reference patterns which are stored in said orthogonalized dictionary means and whose categories are known; converting means for nonlinearly converting the inner products, which are computed by said inner product computing means, in accordance with a positive-negative symmetrical nonlinear function; evaluation value computing means for computing evaluation values on the basis of the nonlinearly converted values by said converting means and coefficients which are preset in units of categories; comparing means for comparing the evaluation values obtained by said evaluation value computing means in units of categories with each other; and discriminating means for discriminating a category to which the input pattern belongs on the basis of comparison results of said comparing means.
Claim 3
Dependent Claim
22 words
A system according to claim 1, characterized in that said evaluation value computing means computes the evaluation values using a neural network.
Claim 4
Dependent Claim
23 words
A system according to claim 1, characterized in that said evaluation value computing means computes the evaluation values using a statistical discriminant function.
Claim 5
Dependent Claim
42 words
A system according to claim 2, characterized in that said neural network has a mulyi-layer structure, and computations at a lowermost layer are replaced with inner product computations between the input pattern and dictionaries orthogonalized by KL expansion in units of categories.
Claim 6
Dependent Claim
110 words
A system according to claim 3, characterized in that said evaluation value computing means computes the following equation on the basis of a mean vector and a covariance matrix obtained in advance from learning data in units of categories, and said discriminating means discriminates a maximum evaluation value as a recognition result using a Bayes decision rule: where µ K and R K are respectively the mean vector and the covariance matrix of an input pattern G obtained from learning data, T indicates transposition, and R K ⁻¹ and | R K | are respectively an inverse matrix of the covariance matrix R K and a value of a determinant.
Claim 7
Dependent Claim
99 words
A system according to claim 3, characterized in that said evaluation value computing means computes the following equation on the basis of a mean vector and a covariance matrix obtained in advance from learning data in units of categories, and said discriminating means discriminates a category having a minimum distance value as a recognition result using a Maharanobis distance: where µK and RK are respectively the mean vector and the covariance matrix of an input pattern G obtained from learning data, T indicates transposition, and R K ⁻¹ is an inverse matrix of the covariance matrix R K .
Claim 8
Dependent Claim
119 words
A pattern recognition method comprising the steps performed by a computer of: computing inner products of an input pattern whose category is unknown and orthogonalized dictionary sets of a plurality of reference patterns whose categories are known; nonlinearly converting the inner products, which are computed in the computing step, in accordance with a positive-negative symmetrical nonlinear function; computing evaluation values on the basis of the nonlinearly converted values by the converting step and coefficients which are preset in units of categories; comparing the evaluation values obtained by the evaluation value computing step in units of categories with each other; and discriminating a category to which the input pattern belongs on the basis of comparison results of the comparing step.
Claim 9
Dependent Claim
119 words
A pattern recognition method comprising the steps performed by a computer of: computing inner products of an input pattern whose category is unknown and orthogonalized dictionary sets of a plurality of reference patterns whose categories are known; nonlinearly converting the inner products, which are computed in the computing step, in accordance with a positive-negative symmetrical nonlinear function; computing evaluation values on the basis of the nonlinearly converted values by the converting step and coefficients which are preset in units of categories; comparing the evaluation values obtained by the evaluation value computing step in units of categories with each other; and discriminating a category to which the input pattern belongs on the basis of comparison results of the comparing step.
Claim 10
Dependent Claim
26 words
A method according to claim 7, characterized in that the evaluation value computing step includes the step of computing the evaluation values using a neural network.
Claim 11
Dependent Claim
27 words
A method according to claim 7, characterized in that the evaluation value computing step includes the step of computing the evaluation values using a statistical discriminant function.
Claim 12
Dependent Claim
42 words
A method according to claim 8, characterized in that said neural network has a multi-layer structure, and computations at a lowermost layer are replaced with inner product computations between the input pattern and dictionaries orthogonalized by KL expansion in units of categories.
Claim 13
Dependent Claim
118 words
A method according to claim 9, characterized in that the evaluation value computing step includes the step of computing the following equation on the basis of a mean vector and a covariance matrix obtained in advance from learning data in units of categories, and the discriminating step includes the step of discriminating a maximum evaluation value as a recognition result using a Bayes decision rule: where µ K and R K are respectively the mean vector and the covariance matrix of an input pattern G obtained from learning data, T indicates transposition, and R K ⁻¹ and | R K | are respectively an inverse matrix of the covariance matrix R K and a value of a determinant.
Claim 14
Dependent Claim
109 words
A method according to claim 9, characterized in that the evaluation value computing step includes the step of computing the following equation on the basis of a mean vector and a covariance matrix obtained in advance from learning data in units of categories, and the discriminating step includes the step of discriminating a category having a minimum distance value as a recognition result using a Maharanobis distance: where µ K and R K are respectively the mean vector and the covariance matrix of an input pattern G obtained from learning data, T indicates transposition, and R K ⁻¹ is an inverse matrix of the covariance matrix R K .
References Cited (Backward Citations)
No citation records indexed.
USPTO Annuity & Maintenance Windows
Maintenance schedule is calculated for granted US utility patents.