Temporal-and spatial-driven video summarization using optimum-path forest

Abstract

Video summarization aims at generating reduced representations for fast and effective video retrieval and classification. In this paper, we cope with such problem by proposing a temporal-and spatial-driven approach that makes use of the Optimum-Path Forest (OPF) clustering to automatic find the number of keyframes, as well as to extract them to compose the final summary. The experiments in two public datasets show OPF can outperform very recent results, thus achieving a performance comparable to some state-of-the-art techniques.