CLASSIFICATION METHOD, CLASSIFICATION APPARATUS, AND CLASSIFICATION PROGRAM

摘要

PROBLEM TO BE SOLVED: To highly accurately perform similarity search using feature quantity vectors.SOLUTION: A generalized inverse stereographic projection parameter setting unit 111 sets the parameters related to the position of a projective point and the position of a hypersphere to be used when generalized inverse stereographic projection is performed. A generalized inverse stereographic projection unit 112a performs inverse stereographic projection onto an m-dimensional feature quantity space V to associate an (m+g-1)-dimensional hypersphere S embedded in a space W, which is g dimensions higher than the m dimensions, with the feature quantity space V. A hyperplane arrangement unit 113 arranges n hyperspheres crossing the (m+g-1)-dimensional hypersphere S in the hypersphere S on the basis of the number of hyperplanes n stored in a hyperplane count storage unit 121. A conversion rule generation unit 114 generates information on conversion rules on the basis of the information of normal vectors and the information of offset coordinates. A bit string generation unit 115a generates a bit string on the basis of the coordinates obtained from the generalized inverse stereographic projection unit 112a.