Locating Anchor Shots in Compression Domain Based on Neural Networks

Anchor shots are important elements in news video, and locating them accurately and thoroughly is crucial to parse news video. The paper presents a novel approach, using neural networks, to detect anchor clips. Firstly, a background model is constructed through neural networks learning. Then, the trained neural networks classify frames in news video into two classes, i.e. anchor frames and non-anchor frames. At last, based on repeatability and dispersing of anchor shots on the temporal axis, false declarations in the outputs of neural networks are filtered out by clustering. The evaluation experiments, on nine days of news videos, demonstrate the approach is a fast, effective one, with the recall 98.2% and the accuracy 100%..