Efficient Visual Information Retrieval using Orthogonal MSTs

The notion of interpoint-distance-based graphs has in the past, guided the extension of distributional order-measures to multivariate observations. The concept of minimal spanning tree (MST) was introduced as the key pattern for generalizing the univariate two-sample problem to multivariate observations. Here, the multivariate Wald-Wolfowitz test is further quantified using the enhanced representations of orthogonal MSTs. Their advantages are investigated by comparing the similarity between color distributions in the feature space, using a standard feature extraction technique borrowed computer vision. To demonstrate the performance of the proposed scheme, the application on a diverse collection of images has been systematically studied in a query-by-example visual information retrieval task. Experimental results show that a powerful measure of similarity can emerge from the statistical comparison of their efficiently drawn pattern representations