To extract each feature axis comprising independent components which are independent of other components from a teaching sample.

Independent components to the number of p which are estimated values of independent signal sources are acquired by independent component analysis or the like from a set of a characteristic distribution of a sample pertinent to a past example, and a feature space wherein p-dimensional independent component vectors based on the acquired independent components to the number of p are used as each feature axis is set. Hereby, overlapping between each feature space of the pertinent sample and a non-pertinent sample is prevented. Moreover, a feature quantity can be set objectively based on a characteristic distribution of an objective sample by determining the feature quantity of the objective sample by an inner product between the characteristic distribution of the objective sample and the independent feature axis.

COPYRIGHT: (C)2007,JPO&INPIT