US Patent 9773023 Image selection using automatically generated semantic metadata

A computer-implemented method for method for selecting visual subject matter stored in a database includes receiving user input that indicates a first region of a first image that is stored in the database and, based on metadata associated with the first image stored in the database, determining at least one visual feature in the first region of the first image stored in the database. The method further includes, based on metadata associated with other images stored in the database, selecting a second image from the database that includes a visual feature that correlates with the visual feature in the first region of the first image, wherein the metadata associated with the first image stored in the database and the metadata associated with the other images stored in the database include image correlation information received from a remote server device.