Mass-based Similarity Measure: An Effective Alternative to Distance-based Similarity Measures

This paper introduces a unique similarity measure that does not compute distance. Instead, it is based on the cardinality of the smallest local region covering the two instances for which the similarity is measured. Theoretical analysis reveals that it is a generalisation of mass estimation. To show its utility, a new information retrieval system is created based on the new similarity measure. Empirical evaluations demonstrate that the new system has a significantly better retrieval performance than five state-of-the-art systems in image and music retrievals.