Abstract
In this paper, we propose to use feature point matching and mesh-based motion estimation for frame rate up-conversion. Matched local feature points between successive frames in a video sequence are employed to constitute a discrete and sparse motion field. Afterwards, a proximity-constraint marking process regulates the raw correspondence set to a content/motion adaptive nodal set. These correspondences are organized by a Delaunay triangular mesh so as to generate a dense motion field through the affine transformation. The experimental results demonstrate that the proposed scheme outperforms three other popular methods in terms of both subjective and objective evaluations. They also suggest that the proposed method is more suitable for videos containing a complex motion field.
Original language | English |
---|---|
Pages (from-to) | 390-397 |
Journal | Neurocomputing |
Volume | 123 |
DOIs | |
Publication status | Published - 10 Jan 2014 |
Keywords
- Motion-compensated frame interpolation (MCFI)
- Frame rate up-conversion (FRUC)
- Feature matching
- Mesh-based motion estimation