飞人:利用基灵矢量场的图像高保真变形

Flying People: High-fidelity Image Warping via Killing Vector Fields

为了把常态实拍人物图像参考指定姿态变形为跳、飞奔、飞跃的"飞人"效果,提出一种利用基灵矢量场的图像高保真变形方法.该方法利用改进的GrabCut算法提取待变形的目标人物;通过语义骨架匹配算法把人物骨架与姿态模板骨架进行匹配;利用骨架约束基灵矢量场实现图像变形,从而达到实拍常态人物图像变形为跳、飞奔、飞跃等效果.实验结果表明,文中方法能够处理人物图像间的变形,有效地把人物图像参照姿态模板进行变形,效果真实.

For realistically warping one real people image with reference to another template gesture so as to make the stand people jump, hop, or skip, etc., a method of high-fidelity image warping via Killing vector fields（KVF） is proposed, which employs the improved Grab Cut algorithm for object extraction, exploits the semantic skeleton match approach for aligning the skeleton, and makes use of the constrained KVF warping image for jump, hop, or skip, etc. Experiments demonstrate that the presented method could warp the images with less relevance, and high-fidelity warp one stand people with reference to different gesture.