Abstract
In this paper we propose a new dynamic video summarization procedure. For this purpose, we find an economical image representation in terms of edge-like features that facilitates video re-sampling with minimal loss of information. We propose a fast noise tolerant way to extract the representation. We argue against the use of an auxiliary descriptor for the samples and we show how we can re-use the existing Principal Components Analysis (PCA) with the appropriate modifications on the new image representation without sacrificing accuracy.