Classification-based motion vector processing for motion compensated frame interpolation using block boundary distortion metric

This paper presents a new classification-based motion vector processing for motion compensated frame interpolation. We first propose to use a block boundary distortion metric calculated by mean square difference of slope combined with the information of frame difference and angle variance to classify motion vectors for reliability level. Then, we correct unreliable motion vectors by using weighted vector median filter and motion vector re-sampling method based on the classified MV map. The experiment results show that the interpolated frames generated by motion vectors with the proposed motion vector processing method have better both visual quality and PSNR performance than the conventional methods.