A Brief Index for Proximity Searching

A large number of computer tasks can be modeled as the search for an object near a given query. From multimedia databases to learning algorithms, data mining and pattern recognition, the metric space model of proximity and retrieval can be used as a searching paradigm. For metric space indexing the permutation based approach consist in saving the order in which a set of reference objects (the permutants) is seen by every element of the database. Adding up the relative displacements with respect to the query is an excellent predictor of proximity, and is called the Spearman ρ distance. In this paper we show how to represent the permutation as a binary vector, using just one bit for each permutant (instead of log k in the plain representation). Hamming distance can be used then to predict proximity. We tested this approach with many real life metric databases obtaining a recall close to the Spearman ρ (or even better in some examples), and speedup from 2 to 4 times faster.

• Type: Conference
• Year: 2009
• Where: CIARP
• Authors: Eric Sadit Tellez, Edgar Chávez, Antonio Camarena-Ibarrola
• Keywords: CIARP 2009, Metric Space, Metric Space Model, Pattern Recognition, Permutation Based Approach