SUBPIXELGENAUE BEWEGUNGSVEKTORSCHÄTZUNG UND BEWEGUNGSKOMPENSIERTE INTERPOLATION

摘要

Sub-pixel accurate motion estimation from digital video requires interpolation of data. In a symmetrical motion estimator that shifts the neighboring images over complementary fractions (-D/2, D/2) of the vector (D), two such interpolating filters are required. In case of block-matching motion estimation algorithms, the match error of various candidate vectors depends on the quality of the sub-pixel interpolation filter(s). As these filters differ per fractional value of the vector, unintentional preferences for some fractional vector values over other may result. The candidate vectors (D) are split into an integer part (Rnd(D/2)) and a generally non-integer part (D-Rnd(D/2)), as this gives a better accuracy and a lower cost than splitting the candidate vectors (D) exactly conform the temporal position at which the vectors need to be valid.