数据驱动的大规模水面动画合成方法

为了生成大规模且细节丰富的水面动画，提出了一种数据驱动的合成方法。整个合成分为两大部分：基于视频的水面重建和三角网格拼接。首先简化了线性近似的影调恢复形状(SFS)方法，针对满足朗伯表面假设的室外大规模水面进行重建，并根据时间和空间的连续性进行了一系列优化处理。然后提出一种针对三角网格序列的拼接方法，根据距离最小原则对重建的表面序列进行采样，将采样结果融合得到无缝的表面。利用以上方法，对四类水面的视频进行重建与拼接，结果表明，合成出的水面具有较高的真实感，在规模与细节之间取得了较好的平衡，该方法可以方便地生成满足用户需求的大规模水面动画。

Data-driven Synthesization for Large-scale Water Animation

A new framework is presented for synthesizing water animation with large size and richdetails. The main idea of our framework consists of two steps: reconstructing mesh model fromvideos and spanning the mesh by quilting. First we simplify the shape from shading method which islinearly approximated, and use it to reconstruct the outdoor water with large scale which satisfies theLambert surface assumption. The result is optimized with several smooth methods according to theconstraint of consistency in time and space. Then a quilting method is introduced for triangle mesh, inwhich we sample smaller blocks from the constructing mesh based on the principle of distanceminimization and merge the blocks to get the seamless surface. The process is completed with fourkinds of water. Results demonstrate that the synthesized water animation has high sense of reality andbalance the scale with detail very well. This method can conveniently synthesize large-scale wateranimation meeting use′s demand.