Apparatus and method for recognizing unknown patterns

Claim 2

A pattern recognition apparatus, said apparatus comprising: memory means for storing a plurality of reference patterns for each category of a predetermined set of patterns, said reference patterns being orthnormalized with respect to each other; input vector generating means for generating an input vector by extracting a feature from an unknown input pattern; basis generating means coupled to said input vector generating means for generating a plurality of bases by deforming said input vector, said bases being orthonormalized with respect to each other; matrix producing means coupled to said memory means and said bases generating means for producing a matrix for each category of said set of patterns, said matrix being formed from a plurality of data elements derived from said bases generated by said basis generating means and from said reference patterns stored in said memory means; and determining means coupled to said matrix producing means for analyzing each of said matrixes to determine the category to which said input pattern belongs.

Claim 3

A pattern recognition apparatus according to claim 1, further comprising: calculating means coupled to said basis generating means and said matrix generating means for obtaining the maximum eigenvalue of said matrix for each category of said set of patterns.

Claim 4

A pattern recognition apparatus according to claim 2, further comprising: comparing means coupled to said calculating means for comparing each of said maximum eigenvalues obtained from said calculating means to select the maximum value among said maximum eigenvalues said selected maximum value representing the category to which said input pattern belongs.

Claim 5

A pattern recognition apparatus according to claim 3, wherein said comparing means further comprises: selection means for selecting the next maximum eigenvalue below said selected maximum value; means coupled to said selection means for comparing said selected maximum value with a predetermined value; and means for comparing said selected maximum value and said next selected maximum value to determine the difference therebetween.

Claim 6

A pattern recognition apparatus according to claim 1, wherein said basis generating means includes: blurring means for generating a blurred pattern from said input vector; differential means coupled to said blurring means for generating a plurality of differential patterns from said blurred pattern; and orthonormalizing means coupled to said differential means for obtaining a plurality of orthonormalized bases, said bases being orthonormalized with respect to each other.

Claim 7

A pattern recognition apparatus according to claim 1, wherein said basis generating means includes: mask storing means for storing a plurality of mask patterns; masking means coupled to said mask storing means for executing a mask operation between said input vector and each of said mask patterns stored in said mask storing means to generate a plurality of resultant masked patterns; and orthnormalizing means coupled to said masking means for orthnormalizing said resultant mask patterns.

Claim 8

A pattern recognition apparatus according to claim 6 wherein said mask storing means stores a blurring mask pattern corresponding to a Gaussian function.

Claim 9

A pattern recognition apparatus according to claim 1, wherein said matrix generating means includes: projector generating means for generating a first projector corresponding to said bases generated by said basis generating means and a second projector corresponding to said reference patterns, and multiplying means coupled to said projector generating means for multiplying said first and second projectors to provide said data elements for said matrix.

Claim 10

A pattern recognition apparatus according to claim 1, wherein said matrix generating means includes; multiplying means for multiplying each of said bases by each of said reference patterns and obtaining a product for each said multiplication; and summing means coupled to said multiplying means for summing said products to provide said data elements for said matrix.

Claim 11

A pattern recognition apparatus according to claim 2, wherein said calculating means calculates the eigenvalues of said matrixes by a power method.

Claim 12

A pattern recognition apparatus for determining the category to which an unknown input pattern belongs, said apparatus comprising: storing means for storing a plurality of reference patterns for each category of a predetermined set of patterns, said plurality of reference patterns forming a first subspace; basis generating means, for generating a plurality of bases from an input pattern, said bases forming a second subspace; matrix producing means for producing a matrix for each category of said set of patterns, said matrix being formed from a plurality of data elements derived from said bases generated by said basis generating means and from said plurality of reference patterns stored in said storage means; calculating means coupled to said storing means and said basis generating means for calculating a similarity value relating to an angle between said first and said second subspace for each category of said set of patterns; and determining means coupled to said calculating means for determining the category to which said unknown input pattern belongs by comparing each of said calculated similarity values.

Claim 13

A pattern recognition apparatus for determining the category to which an unknown input pattern belongs, said apparatus comprising: storing means for storing a plurality of reference patterns for each category of a predetermined set of patterns, said plurality of reference patterns forming a first subspace; basis generating means, for generating a plurality of bases from an input pattern, said bases forming a second subspace; matrix producing means for producing a matrix for each category of said set of patterns, said matrix being formed from a plurality of data elements derived from said bases generated by said basis generating means and from said plurality of reference patterns stored in said storage means; calculating means coupled to said storing means and said basis generating means for calculating a similarity value relating to an angle between said first and said second subspace for each category of said set of patterns; and determining means coupled to said calculating means for determining the category to which said unknown input pattern belongs by comparing each of said calculated similarity values.

Claim 14

A pattern recognition apparatus according to claim 11 wherein said calculating means includes: eigenvalue calculating means coupled to said matrix producing means for obtaining eigenvalues for each of said matrixes produced by said matrix producing means; and detecting means coupled to said eigenvalue calculating means for detecting the maximum eigenvalue for each matrix.

Claim 15

A pattern recognition apparatus according to claim 12 wherein said determining means compares each of the maximum eigenvalues to select the maximum value among said maximum eigenvalues.

Claim 18

A method for recognizing an unknown input pattern according to claim 14, further comprising the step of: obtaining the maximum eigenvalue of said matrix for each category of said set of patterns.

Claim 19

A method for recognizing an unknown input pattern according to claim 15, further comprising the step of: comparing each of said maximum eigenvalues to select a maximum value among said maximum eigenvalues to determine the category to which said input pattern belong.

Claim 20

A method for recognizing an unknown input pattern according to claim 14, further comprising the steps of: generating a blurred pattern from said input vector; generating a plurality of differential patterns from said blurred pattern; obtaining a plurality of orthonormalized bases, said bases being orthonormalized with respect to each other.

Claim 21

A method for recognizing an unknown input pattern according to claim 14, further comprising the steps of: storing a plurality of mask patterns; executing a mask operation between said input vector and each of said mask patterns to generate a plurality of resultant mask patterns; and orthonormalizing said resultant masked patterns.

Claim 22

A method for recognizing an unknown input pattern according to claim 14, wherein said step of producing a matrix comprises the steps of: generating a first projector corresponding to said bases generated by said basis generating means and a second projector corresponding to said reference patterns, and multiplying said first and second projectors to provide said data elements for said matrix.

Claim 23

A method for recognizing an unknown input pattern according to claim 14, wherein said step of producing a matrix comprises the steps of: multiplying each of said bases by each of said reference patterns and obtaining a product for each such multiplication; and summing said products to provide said data elements for said matrix.

Claim 26

A method for recognizing an unknown input according to claim 21, wherein said step of calculating a similarity value comprises the steps of: producing a matrix for each category of said set of patterns, said matrix being formed from a plurality of data elements derived from said bases and from said plurality of reference patterns; obtaining eigenvalues for each of said matrixes; and detecting the maximum eigenvalue for each matrix.