Visualseek: a Content-based Image/video Retrieval Tool A. Image Storage and Retrieval Systems

| This project implements a \content-based" im-age/video retrieval system for the World Wide Web. The application , called VisualSEEk, provides a suite of tools with which a person may search for and retrieve images and videos over the Web. The person formulates visual queries by using the VisualSEEk interface tools to illustrate salient visual properties of the images desired. They query is sent to the server which nds and retrieves to the user the images that best match the visual description in the query. For the scope of the initial project the usable image features include image color and the spatial layout of color regions. For example, the features are used as follows: to retrieve images of \sunsets" from the image archive, one possible query can be constructed by sketching a yellow circle near the center of the image (for the sun) and lling the upper part of the image with orange (for the sky). The images that best match this query will contain colored regions that closely match the speciied color regions in terms of color matches and spatial regions, which should include \sunset" images. The VisualSEEk interface also provides tools to assign other visual properties to query elements. These include texture, color, shape and motion. However, the server application does not yet support these images features. However, the presence of these tools in the interface represents the next direction of enhancement for the VisualSEEk system. The project builds upon the previous Web-based image retrieval system developed at the Columbia University Image and Advanced Television Lab. That application was implemented using the HTML language. It provided for only limited functionality in the user-interface. The system allowed only image searches based on a single color speciied by the user SC95]. The VisualSEEk system instead is implemented in the Java language and shows great improvement in operation. The interface allows for multiple modalities of visual features of the images to be selected by the user. VisualSEEk also utilizes a more powerful search engine that accommodates multiple color regions and their spatial layout. These revisions combine to greatly improve the image retrieval system operation and performance. I. Objective T H e objective of this project is to build a Web-resident application that provides a means searching for and retrieving images and videos from an on-line image archive. At present an image archive consisting of approximately 3200 color images is provided by the …