A Novel Method of Dynamic Textures Analysis and Synthesis

Dynamic textures (or temporal textures) are a kind of visual motion pattern, and their dynamics are not trackable. Dynamic textures such as flames, smoke, and water are statistical stationary and exhibit spatial self-similarity and temporal self-similarity. Recently, description of dynamic textures has attracted growing attention. In this paper, a novel method for analyzing and synthesizing dynamic textures is proposed. First, the dynamic textures are modeled by combination of dynamic texture units. Second, the segment parameters and dynamic texture units´ parameters are learnt at the same time. To reduce computational complexity, we discard Fourier coefficients that are under contribution ratio threshold. Our algorithm is general and automatic, and it works well on various types of textures. Experimental results demonstrate that our approach can reconstruct dynamic texture sequences with promising visual quality and fewer coefficients.