Abstract
The ultimate objective of image retrieval research is to improve the user experiences in dealing with images, which happens to be closely related not only to the retrieval accuracy but also to the way he/she interacts with the retrieval tools. In particular, recognizing the subjectivity inherent in the problem leads to more emphasis on active participation of humans. This paper proposes a new user-system interaction model in the context of similarity retrieval of images. In the proposed model, the interaction is pursued to the extreme so that it tries to help users to browse huge image space with ease and efficiency rather than to find a certain images automatically on behalf of users. The system dynamically reconstructs the view reflecting user commands, while the user continuously modifies his/her commands while seeing the constantly changing view. Here the user's command is called a hint to distinguish it from a query, which is also a representation of user intention, but in a more formalized and complete form. Hints include all the intermediary steps of the user intention description process. By reflecting the intermediary steps to the view immediately, users receive feedback information to modify their description. This gradual and evolutionary process has a huge advantage over traditional approaches, especially when the user intention itself is ambiguous, which is often the case in realistic situations. This paper also describes a simple and efficient multi-dimensional feature indexing algorithm as an enabling technology to ensure immediate response. The algorithm transforms multi-dimensional features to one or more scalar values, which are used for restricting the search space. The algorithm proved to be efficient in realistic situations by being tested on the implementation of the proposed model.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have