News Topic Tracking and Summarization Based on Video Duplicate Detection

点击翻译按钮获取中文摘要

In this project, a novel technique is proposed to detect video duplicates (i.e. video segments with the same file footage, scene, background, object, etc.) from a large-scale broadcast video database. This technique is invariant to the viewpoint of cameras. Based on the assumption that video-duplicate constraints normally guarantee a higher relevance quality than textual constraints, an effective and higher-level news topic tracking and summarization technique is proposed by integrating both visual information and textual information.