The number of nearest neighbours to search for (k) is set to 10, 30, and 50. We randomly generate 20,000 data query points for the image database with dimensions between 2 and 16 in steps of 2. As the dimension goes up, the distances between these 20,000 randomly generated query points are more sparsely distributed and the performance results of the query points in high dimensions can be as close as a query outside clustered data objects.
For a clustered database retrieval application where clusters correspond to object classes, it is likely that queries will not be randomly distributed but will probably be close to or inside the cluster regions. For this reason, another 20,000 query points are generated from another set of similar textures which are evenly spread over the regions of the clustered data objects.
The number of nodes accessed to retrieve knn is recorded for every query point and the mean is calculated from different values of k and the dimensions.