A novel 3D motion co-occurrence matrix (MCM) approach to characterise temporal textures

Temporal textures are motion patterns with indeterminate spatial and temporal extent. Any characterisation technique to be effective has to take into consideration both spatial and temporal motion distribution in an exhaustive manner. Moreover, if the spatial and temporal features can be represented in some integral way, redundancy can be removed, resulting in less computational load and more efficient indexing. In this paper, a novel approach to characterise temporal textures is presented using 3D motion co-occurrence matrix (MCM) to capture the spatiotemporal motion distribution in a unified platform. Thus both spatial and temporal motion distribution is utilized exhaustively and efficiently. In order to make the characterisation process more time efficient the proposed method classifies videos by exploiting already available block-based motion vector information. Experimental results demonstrate the ability of the proposed technique to classify a large set of temporal textures with high accuracy.