Analyzing the inner workings of the Signature Quadratic Form Distance

The Signature Quadratic Form Distance is a distance-based similarity measure comparing flexible feature representations, so-called feature signatures. Thus, the distance's applicability is not limited to feature vectors of the same size and structure. Although the Signature Quadratic Form Distance has shown high retrieval performance in terms of effectiveness and efficiency, the inner workings of this distance are still hidden and not obvious. In this paper, we reveal the distance's inherent structure by empirically analyzing the distance computation and examining their inner workings in order to gain a better model understanding and to guide further developments of the Signature Quadratic Form Distance.