About
The aim of the project was to develop a web-based image and video browser for content-based retrieval, integrating research on searching and browsing into a unified interface. The resulting browser allowed users to efficiently browse large image collections, building upon Daniel Heesch’s work on searching and browsing strategies.
The project focused on integrating searching and browsing mechanisms while improving the user interface. The browser seamlessly incorporated features such as relevance feedback search, hierarchical browsing, lateral browsing via the NNk network, temporal browsing, historical browsing, and an image/video viewer.
The system was implemented using a client-server model, centralising the search engine backend and image collection interface. A Java applet-servlet model ensured platform independence, allowing the client to be accessed from a webpage.
Features and Usage:
- Search: Users could perform text- and image-based searches, adjust visual feature weights, and refine queries through relevance feedback.
- Temporal Frame: Displayed keyframes around a selected image or related images in collections without a time dimension.
- Categories: Provided a default categorisation for browsing.
- NNk Network: Allowed exploration of strongly connected images in a network.
- Viewer: Displayed selected images with details and keywords.
- Output: Enabled users to build and manipulate a collection of selected images.
- Settings: Allowed customisation of image collections, search features, and application configurations.
The browser opened with the TRECVID image collection loaded, providing a structured and interactive browsing experience. By using a Java applet-servlet model, the platform-independent client could be loaded from a webpage.
Publication
A May: Web-based Image and Video Navigation Final year project report. Department of Computing, Imperial College London.
