Non-Parametric Motion Activity Analysis for Statistical Retrieval with Partial Query

摘要: We present an original approach for motion-based video retrieval involving partial query. More precisely, we propose a unified statistical framework allowing us to simultaneously extract entities of interest in shots and supply the associated content-based characterization, which can be used satisfy queries. It relies on analysis motion activity sequences based non-parametric probabilistic modeling information. Areas comprising relevant types are extracted from Markovian region-level labeling applied adjacency graph initial block-based partition image. As consequence, given set videos, able construct structured base samples represented by their models activity. The operations is then formulated as Bayesian inference issue using MAP criterion. report different results extraction examples performed composed one hundred samples.