Efficient and cost-effective techniques for browsing and indexing large video databases

We present in this paper a fully automatic content-based approach to organizing and indexing video data. Our methodology involves three steps: Step 1: We segment each video into shots using a Camera-Tracking technique. This process also extracts the feature vector for each shot, which consists of two statistical variances Var(BA) and Var(OA). These values capture how much things are changing in the background and foreground areas of the video shot. Step 2: For each video, We apply a fully automatic method to build a browsing hierarchy using the shots identified in Step 1. Step 3: Using the Var(BA) and Var(OA) values obtained in Step 1, we build an index table to support a variance-based video similarity model. That is, video scenes/shots are retrieved based on given values of Var(BA) and Var(OA) The above three inter-related techniques offer an integrated framework for modeling, browsing, and searching large video databases. Our experimental results indicate that they have many advantages over existing methods