摘要

Numerous social networks, such as Flickr, Picasa, Instagram, and even news agencies, encourage their users to upload content in the form of annotated images. These image collections have become extremely large, containing hundreds of millions of images from millions of users. One consequence is that retrieval of a full set of thematic images, such as ancient ruins in Italy or mountain climbing in Peru, requires multiple collection searches. This is very time-consuming for the user, particularly because these collections have not been developed for distributed system search and typically have no externally available collection description. In this paper, we present a system that can automatically generate an image collection description suitable for distributed search. Our approach enhances the image tag sets collected by the host system for development of a collection description that provides an extended vocabulary to match search query terms. The benefit of using collection descriptions in the image retrieval process is the ability to first select collections that are relevant to the query before retrieval of relevant images from those collections. This two-step process improves query processing efficiency, because irrelevant collections need not be searched.

  • 出版日期2016-3-25