Abstract : This paper proposes an optimal keyframe selection framework for efficiently and perceptually plausibly summarizing hair animation data composed of a large number of strands. Hair strand animations represent each strand as a continuous curve formed by multiple connected particles, and exhibit high-dimensional deformations over time, which makes it difficult to directly apply conventional particle-based or joint-hierarchy-based keyframe selection methods. To address this challenge, we explicitly exploit the connectivity of strands by defining a root-to-tip relative coordinate system and introducing a Root–Tip structural weighting scheme that assigns higher perceptual importance to regions closer to the strand tip. All strands are normalized into a fixed-dimensional representation via arc-length-based resampling, and global translation components are removed using the mean position of strand roots. This enables a stable high-dimensional vector representation for each frame. Based on the resulting frame representation, we propose a dynamic programming-based globally optimal keyframe selection algorithm that uses segment approximation error as its cost function. The cost combines reconstruction error induced by linear interpolation between keyframes and, optionally, a curvature-based regularization term, so that short yet perceptually important deformations are not missed. Experimental results on diverse hair strand animation sequences demonstrate that the proposed method effectively preserves global shape changes and the dynamic behavior of tip regions even under high frame reduction ratios. In particular, keyframe distribution analysis and position error visualizations confirm that, compared to uniform sampling, our method concentrates keyframes in perceptually important temporal intervals, thereby achieving an effective balance between visual fidelity and data reduction. The proposed framework can be used as a practical keyframe-based summarization approach for storage, transmission, preview, and reuse of hair animations.

[paper]