Content-based Video Parsing and Querying

The recent proliferation of multimedia resources has created a substantial demand for video editing, searching and browsing tools. Our project will focus on developing a user-friendly tool for parsing and querying video on top of Rivl, a resolution independent video language. We will first implement algorithms that compute a short sequence of key frames (a so-called video skim)directly from semi-compressed MPEG video. The user can then select one of the key frames in the skimmed sequence and query for similar key frames. A (rather ambitious) second goal of this project is to develop techniques that allow the user to query for all key frames that contain a particular moving object. We plan to achieve the retrieval of shots with similar scenery and objects with recently developed techniques for image comparison like Color Coherent Vectors and motion tracking algorithms.

People

Comments? Send us mail!

Materials

CS631 home page