Abstract:

The Bounded Coordinate System (BCS) approach is a novel video representation model, which can summarize a video to a coordinate system determined by Principal Component Analysis (PCA). The similarity measure of BCS considers the operations of rotation and scaling for coordinate system matching. However, the BCS approach can not always obtain the minimum distance of rotation and scaling between two bounded Principal Components. To overcome the handicap of the original BCS approach, an improving approach applying dynamic distance measure for rotation and scaling distance is presented in this paper. The new approach can calculate the minimum rotation and scaling distance between two bounded Principal Components, and improve the query precision. Our extensive performance study on a large database of more than  30000 video clips demonstrates that the new approach achieves a higher search accuracy than the BCS approach.

Key words: video search, high-dimensional indexing, principal component, dynamic distance measure