摘要 |
PROBLEM TO BE SOLVED: To make it possible to construct an image dictionary for use in obtaining an image representation from which a characteristic area of an image can be discovered.SOLUTION: A partial area division unit 30 divides an inputted image into partial areas. A classification unit 34 classifies all the partial areas into two sets D1 and D2, and classifies the partial areas of the set of partial areas D1 into groups. A conversion unit 36 performs two-dimensional Fourier transform on feature quantities of the partial areas so as to acquire a transformed partial area expressing each of frequency components of the feature quantities of the partial areas. A discriminator learning unit 38 regards as a positive example the frequency component expressed by the transformed partial area of D1 classified into any of the groups, regards as a negative example, the frequency component expressed by the transformed partial area of D1 that is not classified into any group, learns a first discriminator, performs two-dimensional inverse Fourier transform on the first discriminator learned for each of the frequency components, obtains a second discriminator for each of the groups, and provides an output as an image dictionary.SELECTED DRAWING: Figure 1 |