Approximating adaptive distance measures using scalable feature signatures

摘要

The feature signatures in connection with the adaptive distance measures have become a respected similarity model for effective multimedia retrieval. However, the efficiency of the model is still a challenging task because the adaptive distance measures have at least quadratic time complexity according to the number of tuples in feature signatures. In order to reduce the number of tuples in feature signatures, we introduce the scalable feature signatures, a new formal framework enabling definition of new methods based on agglomerative hierarchical clustering. We show the framework can be used to express nontrivial feature signature reduction techniques including also popular agglomerative hierarchical clustering techniques. We experimentally demonstrate our new feature signature reduction techniques can be used to implement order of magnitude faster yet effective filter distances approximating the original adaptive distance measures. We also show the filter distances using our new feature signature reduction techniques can compete or even outperform the filter distances based on the related feature signature reduction techniques.