Content-based video retrieval using mosaic images

Abstract

Content-based video retrieval is one of the most active multimedia research issues. We propose a system for retrieving video data by means of mosaic images. Mosaicing is a technique of joining/merging frames of a shot into one blended image. The resulting mosaic image can convey three primitive features in videos, that is, background, (moving) objects, and camera work such as pan, tilt and zoom. Thanks to this capability, the user can retrieve video databases effectively by specifying queries encompassing the condition of these three types of video features.