Fast k-nn image search with self-organizing maps

Abstract

Feature-based similarity retrieval became an important research issue in image database systems. The features of image data are useful in image discrimination. In this paper, we propose a fast k-Nearest Neighbor (k-NN) search algorithm for images clustered by the Self-Organizing Maps algorithm. Self-Organizing Maps (SOM) algorithm maps feature vectors from high dimensional feature space onto a two-dimensional space. The mapping preserves the topology (similarity) of the feature vectors by clustering mutually similar feature vectors in neighboring nodes (clusters). Our k-NN search algorithm utilizes the characteristics of these clusters to reduce the search space and thus speed up the search for exact k-NN answer images to a given query image. We conducted several experiments to evaluate the performance of the proposed algorithm using color feature vectors and obtained promising results.