A similarity measure for color image retrieval and indexing based on the Multivariate Two Sample Problem

In this work, a similarity measure in the feature space is proposed for color retrieval and indexing based on the “Multivariate Two-Sample Problem”. Color information is extracted via random selection of image pixels from high-density regions. The proposed scheme has a global nature due to its randomness and is easy to implement. It makes uses of the minimal spanning tree (MST) structure and properties, providing the retrieval results with a statistical measure of their significance level. The main advantages of our proposal are its computational efficiency and the fact that it is generally applicable to natural image collections.