Robust Video Retrieval with Luminance Field Trace Indexing and Geometry Matching

Efficient indexing and robust retrieval are key features for an effective video retrieval system. In this paper we represent video sequences as traces in an appropriate low dimensional space via luminance field scaling and PCA projection, and introduce a combination of top-down and bottom-up strategies for indexing. Various means of introducing distortions in query clips are considered and several heuristics are developed to achieve robustness in retrieval performance. Simulation results demonstrate the effectiveness of the proposed solutions.