TV program segmentation using text-visual analysis

In this paper, we propose a method for detecting semantic segments in a live TV program using closed captions and visual features. Since the unseen closed captions and video frames become available as time passes in live broadcasting, we first divide the currently available video into two groups by computing the segmentation costs, which are the linear combination of the textual and visual segmentation costs. Then, we discover the segment boundaries by computing the stabilities of the segmentations with additional data feed. Experimental results show that the proposed method outperforms the previous work on both precision and recall while processing live TV programs in real-time.