The technology is a new model of user interface software for Content-Based Image Retrieval (CBIR) and digital image organization. Content-based image retrieval systems are systems that retrieve images in the database by visual similarity to query examples (images) specified by the users.
This invention serves three purposes. It improves usability and search performance of CBIR in terms of Graphical User Interfaces; it eases text annotation on a large number of digital images; and it improves the usability of digital image organization.
All transactions are done by creating and manipulating image groups on the workspace. The workspace is divided into two panes. The left pane is the ResultView that displays the result of image search. The thumbnails of search results are tiled on the pane. The right pane is the GroupPalette, where the user manages each image and image groups. In order to create an image group, the user first drags images (thumbnails) of interest from the ResultView to GroupPalette. The user then creates an image group by drawing a rectangle or a circle surrounding these images with a mouse. The user can create any number of image groups.
The Query-by-Group method is an image search method that uses image groups as units of the user's interest. The system retrieves images that are similar to images in the image groups. Meanwhile, the user can also specify some groups as negative or neutral. If a group is specified as a negative, the retrieval system avoids images similar to those negative groups. On the other hand, when a group is specified as neutral, the group does not contribute to retrieval at the moment. The group can be used for another query later.
The user can move images between groups on the workspace by dragging them with a mouse. Therefore, it is easy to try different combinations of query examples. Additionally, it is possible to create a new group within an existing group. Thus, this method achieves greater flexibility of content-based image retrieval such as quick experimental (trial-and-error) query and hierarchical query.
This technology is available in the Open Source. To download, please visit http://www.ifp.uiuc.edu/grouper/