Dimensional Reduction Based on Independent Component Analysis for Content Based Image Retrieval

We propose a novel way to apply independent component analysis (ICA) on eight kinds of visual descriptors (features), and combine the eight features of the same database to extract independent component (IC) feature of each feature. A comparative study on the retrieval performance has been done between the original features and IC features in four image databases. Experiment results show that the IC features are of much less in dimension than the original features, and achieve satisfying retrieval results, sometimes even better than the original results. In this way, hardware storage can be saved in the retrieval preprocessing step.