Adaptively Meshed Video Stabilization

Video stabilization is essential for improving the visual quality of shaky videos. Current video methods usually take feature trajectories in background to estimate one global transformation matrix or several matrices based on a fixed mesh, and warp frames into their stabilized views. However, these may not model camera motion well complicated scenes, such as scenes containing large foreground objects strong parallax, result notable artifacts To resolve above issues, this paper proposes an adaptively meshed method stabilize all its adaptive blocking strategy. More specifically, we first extract then generate triangle mesh according distribution each frame. Then, transformations between views over triangular grids are calculated video. Since more can be extracted from regions, including both finer will obtained provided estimation frame warping. We mesh-based by solving two-stage optimization problem. Moreover, no longer distinguished contribute proposed problem, yielding better performance than previous works, particularly challenging videos with parallax. further enhance robustness our method, propose two weighting mechanisms improve spatial temporal adaptability. Experimental results demonstrate effectiveness producing visually pleasing effects various