Pattern recognizing method and apparatus

A pattern recognizing method for selecting at least one category of pattern recognition candidates for an input pattern, the at least one category being selected from plural such categories of pattern recognition candidates, the method comprising the steps of:

• extracting a feature vector of an input pattern;
• converting the feature vector into a converted feature vector with a conversion matrix, wherein the conversion matrix is chosen such that it converts an intra-mean covariance matrix into a unit matrix, with the intra-mean covariance matrix determined from plural covariance matrices, each of the plural covariance matrices representing covariance between the pattern recognition candidates in one of the plural categories of pattern recognition candidates and a mean vector for that category;
• determining an inner product between the converted feature vector and plural basic vectors, wherein each basic vector is a proper vector of the intra-mean covariance matrix determined when choosing the conversion matrix;
• and
• selecting the at least one category of pattern recognition candidates from a category table on the basis of the inner product, wherein the category table indicates whether regions of a hyperplane defined by the basic vectors overlap onto categories converted by the conversion matrix.

A pattern recognizing apparatus for selecting at least one category of pattern recognition candidates for an unknown input pattern, the at least one category being selected from plural such categories of pattern recognition candidates, the apparatus comprising:

• feature vector input means for inputting a feature vector of the unknown input pattern;
• converting means for converting the feature vector into a converted feature vector with a conversion matrix, wherein the conversion matrix is chosen such that it converts an intra-mean covariance matrix into a unit matrix, with the intra-mean covariance matrix determined from plural covariance matrices, each of the covariance matrices representing covariance between the pattern recognition candidates in one of the plural categories of pattern recognition candidates and a mean vector for that category;
• classification processing means for determining an inner product between the converted feature vector and plural basic vectors, wherein each basic vector is a proper vector of the intra-mean covariance matrix determined when choosing the conversion matrix;
• and
• category selecting means for selecting the at least one category of pattern recognition candidates from a category table on the basis of the inner product, wherein the category table indicates whether regions of a hyperplane defined by the basic vectors overlap onto categories converted by the conversion matrix.

A pattern recognizing method for selecting at least one category of pattern recognition candidates for an unknown input pattern, the at least one category being selected from plural such categories of pattern recognition candidates, the method comprising the steps of:

• inputting a feature vector of the unknown input pattern;
• converting the feature vector into a converted feature vector with a conversion matrix, wherein the conversion matrix is chosen such that it converts an intra-mean covariance matrix into a unit matrix, with the intra-mean covariance matrix determined from plural covariance matrices, each of the covariance matrices representing covariance between the pattern recognition candidates in one of the plural categories of pattern recognition candidates and a mean vector for that category;
• determining an inner product between the converted feature vector and plural basic vectors, wherein each vector for the converted feature vector performed by said converting means, the basic vector is a proper vector of the intra-mean covariance matrix determined when choosing the conversion matrix;
• and
• selecting the at least one category of pattern recognition candidates from a category table on the basis of the inner product, wherein the category table indicates whether regions of a hyperplane defined by the basic vectors overlap onto categories converted by the conversion matrix.