Redundancy Removing by Adaptive Acceleration and Event Clustering for Video Summarization

Abstract

In this paper, we propose a novel approach to summarize rushes. Our processing is composed of several steps. First, we remove unusable content and we dynamically accelerate video according to motion activity to maximize the content per time unit. Then, one-second video segments are clustered into similarity clusters. The most important nonredundant pieces of shot are selected such that they maximize the coverage of those similarity clusters. The produced summaries have been evaluated by an automatic method with a strong positive correlation with the TRECVID campaign evaluation.