Currently available index structures only allow indexing of low-dimensional vectors. However, most index structures available perform filtering instead of ranking, as needed in visual information retrieval. Moreover, it is often necessary to change similarity functions at run time, by attributing different weights to each dimension.