Work

Video Stabilization Algorithm Based on View Boundary Synthesis

Year: 2025

Type: article

Abstract: Video stabilization is a critical technology for enhancing visual content quality in dynamic shooting scenarios, especially with the widespread adoption of mobile photography devices and Unmanned Aerial Vehicle (UAV) platforms. While traditional digital stabilization algorithms can improve frame stability by modeling global motion trajectories, they often suffer from excessive cropping or boundary distortion, leading to a significant loss of valid image regions. To address this persistent challenge, we propose the View Out-boundary Synthesis Algorithm (VOSA), a symmetry-aware spatio-temporal consistency framework. By leveraging rotational and translational symmetry principles in motion dynamics, VOSA realizes optical flow field extrapolation through an encoder–decoder architecture and an iterative boundary extension strategy. Experimental results demonstrate that VOSA enhances conventional stabilization by increasing content retention by 6.3% while maintaining a 0.943 distortion score, outperforming mainstream methods in dynamic environments. The symmetry-informed design resolves stability–content conflicts and outperforms mainstream methods in dynamic environments, establishing a new paradigm for full-frame stabilization.

Source: Symmetry

DOI: 10.3390/sym17081351