Abstract
A variety of smells can be realized by blending multiple odor components using an olfactory display. Since a set of odor components to cover the entire range of smells has not yet been known, we studied a method of selecting odor components using a large-scale mass spectrum database. Basis vectors corresponding to odor components were extracted by the NMF (nonnegative matrix factorization) method. Then, the recipe of the target odor was obtained using the nonnegative least-squares method. The basis vectors were successfully obtained from 10,000 compounds within a tolerable error. Moreover, the mass spectra of 104 odors composed of 322 compounds could be approximated using 32-50 basis vectors.