Segmentation of broadcast news videos by topic

In this paper, a novel system for segmentation of broadcast news program into individual news was designed and implemented. The system consists of several layers. First layer analyzes frames based on the change ratios in their color histogram and edge pixels. Sequential similar frames are marked as a same scene. Next layer aims to find anchors in scene that are marked by the first layer. If anchor is found in N frames, then in order to increase the speed of the analysis, the system stops searching. Third layer analyzes news title regions in the frames in order to understand if two different scenes are related to the same news or not. If two scenes have similar characteristics they are merged into the same scene. Last layer merges anchor scenes with non-anchor scenes to form final segmented news scene of the news program. Scenes with common properties are merged into different layers. System performance was evaluated on broadcast news from different TV channels.