Improved motion vector scaling technique based on half-pixel offset compensation

It is observed that there exists a half-pixel offset in vertical direction between top and bottom fields in interlaced sequence, and this half-pixel offset has an impact on performing motion vector scaling in multipicture motion-compensated prediction. However, the impact caused by this half-pixel offset is processed inconsiderately during motion vector derivation in the up-to-date video coding standards, where temporal motion vector predictor usually derives its forward or backward motion vector by performing motion vector scaling based on temporal distance. In this paper, an improved motion vector scaling technique is proposed to compensate the impact caused by the half-pixel offset between top and bottom fields. Motion vector scaling is performed on corresponding coordinates which represent the position relationship between top and bottom fields properly. Experimental results show that the proposed motion vector scaling technique improves coding efficiency for interlaced sequence, especially for bi-predictive pictures.