After the analysis of B'[m,n] in regard to the above mentioned thresholds the features are extracted from all remaining non-zero labeled regions. For each of these regions the generative color set 
is recorded and the region area and location are measured. The region location is represented by the minimum bounding rectangle (MBR) that encloses the region. The features are added to the feature table which compiles the meta-data to be used to answer the color region queries.
Color indexing using the binary color sets is accomplished by indexing the color region feature table on the color set attribute. Since this field contains a binary vector, a binary tree is an appropriate index structure. In a straight-forward color set query such as find all images that have a region with precisely this color set, the matches are found by looking up the color set value in the index and reading off the images. However, more interesting queries specify several color regions and their spatial layout.
The VisualSEEk system presently supports queries that include up to three color regions and their spatial positions. The system allows the user to select priorities for matching color and spatial positions to the items in the database. VisualSEEk allows the user to specify whether the color matches are to be exact match or best match. Exact match means the color sets must be exactly equal for a match. For best match the system also examines nearby color sets and assigns matches a color distance so that the system returns items in order of color set distance. VisualSEEk also allows the user to specify whether the spatial locations in the query specify are exact or best locations or none. For best match the system assigns a spatial distance depending on the drift from the locations specified in the query. In the case of none the system ignores the location parameters. The system also allows the user to specify the maximum number of images 
to be returned by the query.