Optimising video summaries using unsupervised clustering

Following the paradigms of the video summarisation for production professionals, this work brings an efficient unsupervised method for ranked visual summarisation of large-scale video repositories. Following the pre-processing stage of frame clustering and key-frame extraction, a novel graph based analysis ranks the key-frames based on their saliency in the final video summary. In order to evaluate the ranking process, a comic-like layout of key-frames has been applied. The experimental results show good subjective summarisation of the analysed content, while maintaining notion of the temporal structure and knowledge discovery.