The rise of robust digital techniques has lead to numerous multimedia products in which video plays an important role. Due to the different video formats in these products, Video Format Conversion (VFC) became a key technology. For the temporal interpolation in video conversion, motion estimation (ME) can be used. This motion estimation is often block-based, as this gives a superior price-performance ratio compared to other ME methods. However, motion estimation suffers from the so-called occlusion problems, where differently moving objects in fore- and background cause covering and uncovering (occlusion) of certain regions in an image. In this report the main objective is to reduce these occlusion problems. In finding a solution, it is kept in mind that eventually this solution is to be used in motion compensated upconversion.
In block-based motion estimation, solving occlusion problems can be split in two sub-problems. The first is to find a block-based motion estimation method, which is able to estimate a correct vector field in occlusion areas. The second is to solve blocking artefacts, a specific problem resulting from block-based motion estimation. A solution to the first problem was found n a new 3-frame estimation approach, where the error criterion in the estimator played an important role. Solutions to the second problem were found in estimation on smaller blocks and in-block segmentation.
All evaluated error criteria suffer from problems, especially from the extension of the foreground vector into the background. Thereby, the aperture problem plays an important role, which increases if noise is present. The achieved reduction of blocking artefacts by the use of smaller blocks and segmentation also depends heavily on the noise-level. Smaller blocks suffer more from the aperture problem and noise makes it even worse. Segmentation resulting in small regions suffers from the same problems.
The 3-frame approach definitely decreases
the occlusion problems, compared to former 2-frame approaches. This can
be seen from a comparison with a new upconversion approach. However, further
research is required to design a good performing 3-frame error criterion.
The reduction of blocking artefacts also contributes to the reduction of
occlusion. Especially when noise is present, the use of smaller blocks
should be preferred over segmentation as it is computationally cheaper.