Abstract

Existing video stabilization methods either require aggressive cropping offrame boundaries or generate distortion artifacts on the stabilized frames. Inthis work, we present an algorithm for full-frame video stabilization by firstestimating dense warp fields. Full-frame stabilized frames can then besynthesized by fusing warped contents from neighboring frames. The coretechnical novelty lies in our learning-based hybrid-space fusion thatalleviates artifacts caused by optical flow inaccuracy and fast-moving objects.We validate the effectiveness of our method on the NUS and selfie videodatasets. Extensive experiment results demonstrate the merits of our approachover prior video stabilization methods.

 

Quick Read (beta)