A Video Mosaicking Technique with Self Scene Segmentation for Video Indexing

It is important to extract key information from video for the purpose of indexing and fast scene retrieval. Conventional frame-based video representation is appropriate for viewing it in a movie mode, but is not adequate for efficient access to information of interest. Therefore, Scene-based video representation using an image mosaicking for video indexing has been proposed recently. The scene segmentation is the first step of an image mosaicking because a mosaic image is composed of background of all frames that comprise the scene. Therefore, the image mosaicking with simultaneous scene segmentation is the natural choice for an efficient video representation. In this paper, we present an image mosaicking algorithm with efficient and robust automatic scene segmentation using phase correlation and motion-based algorithm. Simulation results show that the proposed method is fast and robust for the scene change detection and appropriate for scene-based video indexing.