ITERATIVE INCREMENTAL SHOT CLUSTERING ALGORITHM BY HAAR WAVELETS

摘要

Video shot clustering is the basis of other high-level researchrnof multimedia databases applications. This articlernproposes a novel and efficient shot clustering algorithm forrnvideos by applying the multi-resolution analysis of Haarrnwavelets which is called MLHC(Multi-Level HierarchicalrnClustering). Corresponding to the reconstruction proceduresrnof Haar wavelets, MLHC is designed as a multi-levelrnalgorithm. When the algorithm runs to further levels, thernclustering results are increasingly credible and precise. Afterrnthe clustering results achieve a stable status, MLHCrnstops automatically. Thus it’s an iterative incremental clusteringrnalgorithm. Each level of MLHC is an independentrnhierarchical clustering algorithm which resolves therndilemma of choosing proper initial cluster centers for mostrnexisting shot clustering algorithms. For each hierarchicalrnlevel of MLHC, a novel stop criterion is designed to stoprnthe iterative merging procedures and terminates MLHC onrnthis level. By this stop criterion, the clustering results canrnbe obtained automatically without any parameters and thernnumber of clusters can also be estimated at the same time.rnThe theoretical analysis and the extensive experiments witnessrnthe efficiency and effectiveness of our proposals.