Matching Based On Feature Points, Electronic Image Stabilization Algorithm Research

Digital video stabilization technology aims to eliminate the jitter of video content, providing an easy access to observe dithering video sequence, which is comprised by three key models, named motion estimation, motion filtering and motion compensation. Motion estimation is applied to estimate the global motion parameter between frames. It should be noted that the motion estimation (ME) in video stabilization context is different to the ME model in video coding counterpart, which distinguish itself by its high requirement of estimating motion parameter in real time from two frames that intensive movement may be exhibited in. To extract the intentional motion from estimated motion parameters, motion filtering is employed to remove the unwanted jitter. Finally, each frame is adjusted to an appropriated position through motion compensation using the modified motion parameter.We have proposed a global motion estimation method based on parameter space voting and edge points matching, called PVEPM technology, which is comprised by several steps. In the first place, image are pre-processed by exist filtering methods, and then edge points are extracted. In the following steps, some new method such as delusion filter and restricted search algorithm are proposed, which are the innovation point of this thesis. Experiment proves that our algorithm has great advantages on the robustness and computation efficiency. Test under various image scenes also indicates that this method is capable of tolerating high noise.At end of this thesis, motion filtering and motion compensation technology are also addressed.