Pattern recognition system using posterior probabilities

Claim 2

A pattern recognition system comprising: feature extracting means for extracting a feature of an input pattern; memory means for storing a respective reference pattern for categories of patterns to be recognized; similarity calculating means for calculating a similarity value between the feature extracted by said feature extracting means and one of the reference patterns stored in said memory means; posterior probability transforming means for transforming the similarity value calculated by said similarity calculating means into a posterior probability, wherein said posterior probability transforming means includes means for calculating the posterior probability by using a parameter set required for calculating the posterior probability on the basis of the similarity value calculated by said similarity calculating means, said parameter set obtained in advance via recognition processing from training similarity value data; and means for recognizing said input pattern on the basis of said posterior probability.

Claim 3

A pattern recognition system comprising: feature extracting means for extracting a feature of an input pattern; memory means for storing a respective reference pattern for categories of patterns to be recognized; similarity calculating means for calculating similarity values between the feature extracted by said feature extracting means and the reference pattern stored in said memory means; posterior probability transforming means for transforming at least one of the similarity values calculated by said similarity calculating means into a posterior probability, wherein said posterior probability transforming means includes: transformation parameter memory means for storing, for each of the categories, a parameter set including parameters for defining a distribution of similarity values correctly recognized in recognition processing derived in advance from training similarity value data, parameters for defining a distribution of similarity values erroneously recognized in the recognition processing, and a weighting coefficient required for calculating the posterior probability from the two similarity value distributions; calculating means for calculating the posterior probability on the basis of the similarity values calculated by said similarity calculating means and the parameter set stored in said transformation parameter memory means; and means for recognizing said input pattern on the basis of said posterior probability.

Claim 4

A pattern recognition system comprising: feature extracting means for extracting a feature of an input pattern; memory means for storing a respective reference pattern for categories of patterns to be recognized; similarity calculating means for calculating similarity values between the feature extracted by said feature extracting means and the reference pattern stored in said memory means; posterior probability transforming means for transforming at least one of the similarity values calculated by said similarity calculating means into a posterior probability, wherein said posterior probability transforming means includes: transformation parameter memory means for storing, for each of the categories, a parameter set including parameters for defining a distribution of similarity values correctly recognized in recognition processing derived in advance from training similarity value data, parameters for defining a distribution of similarity values erroneously recognized in the recognition processing, and a weighting coefficient required for calculating the posterior probability from the two similarity value distributions; calculating means for calculating the posterior probability on the basis of the similarity values calculated by said similarity calculating means and the parameter set stored in said transformation parameter memory means; and means for recognizing said input pattern on the basis of said posterior probability.

Claim 5

A system according to claim 2, wherein said transformation parameter memory means includes means for storing, for each of the categories, a parameter set including parameters α and β for defining the distribution of similarity values correctly recognized in the recognition processing derived in advance from the training similarity value data, parameters α and β for defining the distribution of similarity values erroneously recognized in the recognition processing, and a weighting coefficient ω required for calculating the posterior probability from the two similarity value distributions.

Claim 6

A system according to claim 3, wherein said calculating means includes means for transforming a similarity into the following equations on the basis of the similarity value calculated by said similarity calculating means and the parameter set including α, β, α, β and ω stored in said transformation parameter memory means: ##EQU12## and calculating the posterior probability by the following transformation equation: ##EQU13## wherein: i is an index; Ti is an event; wi is an input pattern; and si is a similarity value.

Claim 7

A pattern recognition system comprising: feature extracting means for extracting a feature of an input pattern; memory means for storing a respective reference pattern for categories of patterns to be recognized; similarity calculating means for calculating a similarity value between the feature extracted by said feature extracting means and one of the reference patterns stored in said memory means; posterior probability transforming means for transforming the similarity value calculated by said similarity calculating means into a posterior probability, wherein said posterior probability transforming means includes table memory means for storing a transformation table for calculating the posterior probability calculated by using a parameter set including a parameter for defining a distribution of similarity values correctly recognized in recognition processing derived in advance from training similarity value data with respect to the reference pattern of each category on the basis of the similarity value calculated by said similarity calculating means, a parameter for defining a distribution of similarity values erroneously recognized in the recognition processing, a weighting coefficient required for calculating the posterior probability from the two similarity value distributions, and the similarity value calculated by said similarity calculating means; and means for recognizing the input pattern on the basis of the posterior probability.

Claim 8

A pattern recognition system comprising: feature extracting means for extracting a feature of an input pattern; memory means for storing a respective reference pattern for categories of patterns to be recognized; similarity calculating means for calculating a similarity value between the feature extracted by said feature extracting means and one of the reference patterns stored in said memory means; posterior probability transforming means for transforming the similarity value calculated by said similarity calculating means into a posterior probability, wherein said posterior probability transforming means includes table memory means for storing a transformation table for calculating the posterior probability calculated by using a parameter set including a parameter for defining a distribution of similarity values correctly recognized in recognition processing derived in advance from training similarity value data with respect to the reference pattern of each category on the basis of the similarity value calculated by said similarity calculating means, a parameter for defining a distribution of similarity values erroneously recognized in the recognition processing, a weighting coefficient required for calculating the posterior probability from the two similarity value distributions, and the similarity value calculated by said similarity calculating means; and means for recognizing the input pattern on the basis of the posterior probability.

Claim 9

A system according to claim 5, wherein said table memory means includes means for storing, for each of the categories, a posterior probability calculated by using the parameter set including parameters, α and β for defining the distribution of similarity values correctly recognized in the recognition processing derived from training similarity value data of each category, parameters α and β for defining the distribution of similarity values erroneously recognized in the recognition processing, a weighting coefficient ω required for calculating the posterior probability from the two similarity value distributions, and the similarity value calculated by said similarity calculating means.

Claim 10

A system according to claim 6, wherein said table memory means includes means for transforming a similarity into the following equations on the basis of the similarity value calculated by said similarity calculating means and the parameter set including α, β, α, β and ω stored in said transformation parameter memory means: ##EQU14## and storing the posterior probability calculated by the following transformation equation: ##EQU15## wherein: i is an index wi is an input pattern; and si is an similarity value.