A New Update Strategy for Blocks with Low Correlation in 3-D Recursive Search
Wontae Kim, Sehun Kim, Jin-Sung Kim, Hyuk-Jae Lee · 2018
The conventional 3-D recursive search (3DRS) often fails in the derivation of an accuracy motion vector when the motion of a block is quite different from those of spatially and temporally adjacent blocks. In order to find a motion vector of un uncorrelated block by using a 3DRS, this paper proposes a new strategy of motion vector update in which spatial and temporal correlations are not used to generate update candidates. In addition, an optimization for reducing additional computation for the new update strategy is proposed. The proposed algorithm aims to derive a true motion vector not for all blocks, but for at least a single block in an object, which reduces the number of blocks to be processed. Then, the true motion vector is propagated to the other blocks in the object. As a result, the motion vectors of uncorrelated blocks are derived by the proposed algorithm without a significant increase of the number of update candidates per a block. Experimental results show that the proposed 3DRS significantly improves the accuracy of the motion vector for an uncorrelated block compared with the conventional 3DRS.