Video Classification and Mining Based on Statistical Methods for Cross-Correlation Analysis

In this paper, we present a novel method for statistical cross-correlation for video analysis. We randomly sample the cross-correlation function in order to dramatically reduce the search time for the maximum cross-correlation coefficient. We subsequently develop a method to monitor the likelihood that a significantly higher cross-correlation coefficient value could be extracted based on sequential hypothesis testing. We terminate the search when a threshold on the likelihood has been reached and rely on the largest cross-correlation coefficient sampled for video classification and mining applications. Computer simulation experiments demonstrate the dramatic reduction in speed requirements using the proposed statistical cross-correlation analysis method, while the classification performance remains comparable to the performance achieved using exhaustive search.