| 摘要 |
<p>An apparatus for recognising unknown input patterns is described. The pattern to be recognised is converted into electrical signals representing, for example, the darkness or density of the picture elements forming the input pattern by units 1, 2, 4 and 4. The signals are stored in an input pattern memory 4. A basis generating unit 5 generates a plurality of orthonormalised vectors or bases from the stored input pattern, these are used to construct an input subspace of the input pattern. A reference pattern memory 8 stores reference patterns for a number of categories. These reference patterns vectors which are used to construct a reference subspace. A similarity value is defined as an angle theta between the input subspace and the reference subspace. A matrix generating unit 7 and an eigenvalue calculating unit 10 are provided for measuring the angle theta . The matrix generating unit 7 produces a matrix from the orthonormalised vectors generated by the basis generating unit 5 and the reference patterns stored in the reference memory 8. The maximum eigenvalue of the matrix calculated by eigenvalue calculating unit 10, corresponds to the angle theta . A decision unit 12 determines the category to which the input pattern belongs by comparing the eigenvalues representing the categories. </p> |
