A motion compensated frame interpolation scheme for high-definition videos

This paper presents a motion compensated frame interpolation (MCFI) scheme for high-definition (HD) video applications. The proposed algorithm uses only two successive frames, and the motion candidates are selected from the spatial and temporal domains. The scheme includes the following four steps. Firstly, forward and backward motion fields are estimated. Secondly, the motion vector fields are projected to the interpolated frames by our new retiming method based on median filter. Thirdly, a geometric based occlusion detection algorithm is applied to find the covered and uncovered regions in the interpolated frame for unidirectional interpolation. Finally, new frames are generated by unidirectional and bi-directional interpolation to reduce the halo artifacts. Experimental results demonstrate that the proposed algorithm achieves much better visual quality than the traditional methods.