Video copy recognition by Oriented PCA and statistical analysis

In this paper we first propose a feature model clustering visual features for video copy recognition, and adopt Oriented PCA (OPCA) to compute subspace feature for robustness to video distortions and dimensionality reduction. We also propose a novel method to explore statistics of video database to estimate nearest neighbor classification error rate and learn the optimal classification threshold. Recognition performance is evaluated under significant video distortions and different video length. Results show that recognition error rate below 5% has been achieved under significant distortions, and subspace representation leads to much reduction of error rate compared to using original feature, especially for very short video clips (e.g.5s).