Frame rate up-conversion using segment-based motion estimation

Abstract

In this thesis, we present a new frame rate up-conversion method by using segment-based motion es-timation. By using segment-based motion estimation, the proposed method can obtain more accurate motion vectors than the block-based method, even for some large and homogeneous regions or regions with small objects and thin lines, which usually produce wrong motions when the block-based method is used. The pro-posed method can also give reliable motion vectors even for the segments included in occluded regions, by using motion vector sharing, which shares motion vectors between consecutive motion vector fields for the same segment. In addition, the proposed method can tolerate for some wrongly segmented regions by re-estimation for sub-segment, which is derived by over-segmentation if it is necessary. Finally, the proposed method achieves much better interpolated results by using pixel-wise refinement and some proposed interpo-lation schemes. Experimental results demonstrated that the proposed method outperforms the previous methods, which exploited the block-based method.