Abstract : This paper proposes a perceptually guided keyframe selection framework for efficiently summarizing and compressing large-scale 3D water particle simulation animations. Conventional keyframe selection methods developed for motion capture data assume fixed joint structures and hierarchical representations. In contrast, water particle data exhibit a varying number of particles across frames and lack any inherent structural hierarchy, making such formulations difficult to apply directly. Moreover, directly processing large particle sets in high-dimensional spaces leads to a significant increase in computational cost. To address these challenges, we project particle sets into camera-aligned screen space and perform frontmost particle sampling on a fixed-resolution grid, selecting a single visible particle per cell. This strategy yields a fixed-length, constant-dimensional representation regardless of frame-to-frame variations in particle count. In addition, we define a centroid-relative coordinate system for each frame and apply distance-based importance weighting, enabling stable comparison and approximation cost evaluation even for particle sets without explicit structural reference points. Based on this representation, we apply a dynamic programming-based optimal keyframe selection algorithm to summarize the animation sequence using a small number of representative frames. Experimental results demonstrate that the proposed method effectively preserves dominant flow structures across diverse water simulation scenes while substantially reducing the number of frames. The resulting reconstruction errors are primarily confined to high-velocity and splash regions, indicating that the perceptual fidelity of the animation is largely maintained.

[paper]