Summarization of news videos based on the consistency of news text and video contents

This report proposes a method for summarizing news videos by selecting scenes whose image contents are consistent with the news texts. News videos are valuable resources because daily events are recorded in them. Therefore, there is a demand for efficiently watching a large amount of them. Some text-based methods for summarizing news videos have been proposed. However, they summarize news videos based on only their text contents, which leads to the problem that they do not take into account their image contents. Thus, the proposed method selects scenes whose image contents are consistent with the text contents considering “Visual Concepts” of each scene. Two kinds of subjective experiments were performed in order to evaluate the performance of the proposed method. As a result, we confirmed the effectiveness of the proposed method for various news topics.