News video classification using multimodal classifiers and text-biased combination strategies

Automatic classification of video content is a key technique in video management. Such classification combines evidence from multiple modes, but most existing combination strategies use a unified approach to process multimodal features, neglecting the asymmetric impact of textual and audio-visual features. This paper presents a text-biased scheme integrating multiple modes in news video classifications. The combination strategy mainly depends on textual evidence with some input from audio-visual clues. The classification approach has been validated on large-scale live news videos.