|Title||A Super-resolution Approach using Combined Inverse Wiener and Global Motion Temporal Filtering|
|Tutor||Dr.-Ing. Andreas Krutz|
|Professor||Dr.-Ing. Thomas Sikora|
|Abstract||Size increasing of videos can be achieved by using interpolation techniques, while the improvement of the video sequence demands the image restoration approaches, in which the global-motion-estimation technique plays an important role in this thesis. |
Therefore, choosing a reasonable interpolation approach and a global motion estimation method and combine them together to obtain "super-resolution" video sequences are the most important issues in this work.
A brief introduction of the fundamental theory of the interpolation technique is represented and some different interpolation methods are introduced and compared. The one using an optimal inverse interpolation filter in a Wiener scene is chosen for this work, considering all effects and so to compute efficiency.
The estimation of motion between consecutive images taken from a video sequence is another important issue of this work. A perspective motion model is used to describe the motion features. For an acceptable performance in estimation the Gauss-Newton algorithm (a gradient descent algorithm) has been applied.
Several successive frames are warped using the parameters estimated and filtered on order to implement the quality enhancement. This thesis mainly focuses on the resolution enhancement of the background. The presented super-resolution approach based on combining filtering is then tested, using several video sequences containing different motion types.
|Key words||global motion estmation, camera modeling, noise analysis, image enhancement methods, interpolation, Wiener Filter, Gauss-Newton Algorithm, filtering|