Full-Frame Video Stabilization via SIFT Feature Matching

Removal of visually unpleasant motion from videos is an important video enhancement technology. We present a feature-based approach for video stabilization that produces stabilized videos, while preserving the original resolution. To characterize the global/camera motion, SIFT features are extracted and carefully chosen to define the homography of 2D perspective warp between two consecutive frames. To remove high-frequency components from the original motion path, the time-domain filter is applied to generate the motion-compensated version of transformation associated with each frame. Then, a missing pixel is filled in by the average of the warped pixels. The experimental results have shown that the proposed method is effective in full-frame video stabilization.