An useful method for scene categorization from new video using visual features

The amount of multimedia and broadcasting contents is increasing; on the other hand, modern people want to use selectively and quickly only important part. This propensity to consume makes consumer or service provider need an efficient technology about summary extraction and important scenes detection from whole video. Especially news headlines are very important part that summarize about one hours of news video. So we proposed a method to extract headlines scenes automatically from news video as well as categorize anchorperson scene and reporter scene using multiple MPEG-7 visual features. Firstly multiple features are extracted from key frames which represent each shot divided by the shot-boundary making process. And some shots are classified to news headlines part, anchorperson part, and reporter part by threshold calculating and decision after combination of multiple visual features. Experimental data show the proposed method provides good performance in automatic scene categorization from whole news video. Our method could be applied efficiently in news video highlight, summary service or vod service.