Correspondence learning apparatus and method as well as the corresponding relationship learning program, annotation apparatus and method and program for annotation, and retrieval apparatus and method and program for retrieval

摘要

An image data processing system (20) has a learning storage apparatus (41) that stores projection matrixes (A,B) obtained by canonical correlation analysis so as to derive, based on at least one of an image feature (xi) and a word feature (yi), a latent variable (si) as an abstract concept (L) used for associating an image with a word corresponding thereto and that further stores information required for obtaining the latent variable (si) acquired by use of the projection matrixes (A,B), a probability P(x/si) of occurrence of an arbitrary image feature (x) from a certain latent variable (si) and a probability P(y/si) of occurrence of an arbitrary word feature (y) from a certain latent variable (si). In this way, a probability P(x,y) of the image feature (x) and word feature (y) being simultaneously outputted can be easily and quickly determined, thereby executing a high-speed annotation or retrieval with high precision.