The VisualSEEk project
has emphasized several unique objectives in order to improve image retrieval, such as: (1) automated extraction of localized regions and features [13], (2) querying by both feature and spatial information [14], (3) feature extraction from compressed data [15], (4) development of techniques for fast indexing and retrieval and (5) development of highly functional user tools.
The VisualSEEk client application was developed in the Java language to allow for client platform independence and accessibility on the World Wide Web. As illustrated in Figure 3, the VisualSEEk system consists of several components: the set of user tools, the query server, the image and video server, the image and video archive, the meta-data database and the index files. Currently, the VisualSEEk system allows searching for images in a test-bed of 12,000 color images. The VisualSEEk tools are also being ported to an application for searching in a collection of over one million images and videos from the World-Wide Web [16].
In the next section, we present the color set representation and back-projection region extraction process. In section 3, we examine the process of querying by color similarity. In section 4, we extend the color query process to include the matching of regions that have properties of absolute location, color and size. In section 5, we consider the case of matching images with multiple regions that have properties of color, size, absolute and relative locations, which includes cases of special relations between regions. Finally in section 6, we provide an evaluation and present some examples of querying by joint color/spatial features.