Efficient content-based retrieval: experimental results

Extensive testing has shown that the bare-bones triangle inequality algorithm could be used to sharply reduce the number of images needed to be directly compared to a query image for a given distance measure, and that adding the triangle-trie for a two-stage algorithm can be used to search for matches faster than even the bare-bones triangle inequality algorithm. We have developed a method for using the triangle inequality algorithm for combinations of distance measures, thus allowing for database systems which combine flexibility and speed. There are a number of open problems concerning the various data structures and algorithms we describe: key selection, number of keys, trie depth and bin size. More generally, the statistical behavior of distance measures over different sets of images influences the behavior of all the algorithms, and this needs to be explored