FELICITY: A flexible video similarity search framework using the earth mover’s distance

Abstract

In this paper, we demonstrate our novel system, called FELICITY, which is capable of processing user-adaptive content-based k-nearest-neighbor queries efficiently by utilizing video signatures and the Earth Mover’s Distance (EMD). To this end, we implement an optimal multi-step query processing algorithm by approximating the EMD between any two video signatures by utilizing lower-bounding filter approximation techniques. The system enables the user to adapt query parameters and interact with the system, helping him to find out the best parameter combination for the desired similarity tasks. Moreover, our system incorporates a user-friendly visualization interface for the EMD flow between video signatures, providing an intuitive understanding of the EMD and video similarity.