| 主权项 |
1. A recognition dictionary generating device that generates a recognition dictionary formed by plural reference vectors, comprising:
an acquiring unit that acquires a plurality of (d+1)-dimensional reference vectors (d is an integer not less than 1) each containing d pieces of feature values and an offset value indicating a degree of importance of each of the reference vectors, and a (d+1)-dimensional learning input vector; a selection unit that, from among the plurality of reference vectors acquired by the acquiring unit, selects a first reference vector belonging to a class same as the learning input vector and having a minimum distance from the learning input vector, and a second reference vector belonging to a class different from the learning input vector and having a minimum distance from the learning input vector; a distance acquiring unit that acquires a first distance value indicating a distance between the learning input vector and the first reference vector, and a second distance value indicating a distance between the learning input vector and the second reference vector; a correcting unit that corrects the first reference vector using a first correction vector obtained by multiplying a coefficient changing in accordance with a relationship between the first distance value and the second distance value, by a value obtained by exponentiation of the second distance value, and by a difference between the learning input vector and the first reference vector, and corrects the second reference vector using a second correction vector obtained by multiplying the coefficient, by a value obtained by exponentiation of the first distance value, and by a difference between the learning input vector and the second reference vector; and a determining unit that determines a reference vector to be excluded from the recognition dictionary in accordance with an offset value of each of the first reference vector and the second reference vector, each of which has been corrected by the correcting unit. |
