Automated summarization of narrative video on a semantic level

The movie industry produces thousands of feature films and TV series annually. Such massive data volumes would take consumers more than a lifetime to watch. Therefore, summarization of narrative media, which engages in providing concise and informative video summaries, has become a popular topic of research. However, most of the summarization solutions so far aim to represent just the overall atmosphere of the video at the expense of the story line. In this paper we describe a novel approach for automated creation of summaries for narrative videos. We propose an automated content analysis and summarization framework for creating moving-image summaries. We aim at preserving the story line to the level that users can watch the summary instead of the original content. Our solution is based on textual cues available in subtitles and movie scripts. We extract features like keywords, main characters names and presence, and combine them in an importance function to identify the moments most relevant for preserving the story line. We develop several summarization methods and evaluate the quality of the resulting summaries in terms of user understanding and user satisfaction through a user test.