Content-Based Story Segmentation of News Video by Multimodal Analysis

摘要

This paper proposes a method of story segmentation for news video using multimodal analysis. The method detects the topic-caption frames, integrates them with silence clips detection, shot segmentation to locate news story boundaries. On test data with 135,400 frames, the accuracy rate 87.9% and the recall rate 98.7% are obtained. The experimental results show the method is promising and robust.