Multi-resolution parametric synthesis of manipulative dynamic textures.

Xu, Leilei; Sun, Hanqiu; Wang, Lifeng

Dynamic textures are composed of spatially multi-scale and temporally coherent visual signals. In this paper, we propose a novel approach for multi-resolution parametric synthesis (MPS) of manipulative dynamic textures, including learning, analysis, and motion manipulation of layered feature spaces. The proposed approach separates the motions into different scales or frequencies, and the dynamic feature spaces are analyzed in modeling and then mapped to a global coordinate system to expose the multi-scale structures. Our approach emphasizes the importance of fine-scale data appearing more informative, while coarse signals are less focused. The synthesized dynamic textures using our MPS approach are strongly enhanced in dynamic appearance than the single-scale LDS model, and also flexible in manipulating the decomposed feature spaces for creating different dynamic textures of the real world. Our experimental results showed the more faithful synthesized dynamic textures from input short videos, and their manipulative controls by simple editing of feature space basis, not only the whole video data. Copyright © 2008 John Wiley & Sons, Ltd.

COMPUTER vision; PARAMETRIC devices; PATTERN recognition systems; GRAPHIC methods; NEURAL computers; COMPUTER science education

Computer Animation & Virtual Worlds, 2008, Vol 19, Issue 3/4, p387

1546-4261

Academic Journal

10.1002/cav.240