FIELD: video encoding and decoding.
SUBSTANCE: motion vector is generated for a subblock of the specified block. A motion prediction signal is generated based on the subblocks using the motion vector of the subblock. A set of motion vector difference values associated with the subblock of the current block is determined at the pixel level, wherein the pixel level motion vector difference indicates the difference between the pixel level motion vector obtained from the subblock sample location and the subblock motion vector. Subblock spatial gradients are determined based on the motion prediction signal at the location of each subblock sample. A motion prediction refinement signal is determined for the current block based on a set of motion vector difference values and determined spatial gradients determined at the pixel level. The subblock-based motion prediction signal and the motion prediction refinement signal are combined to generate a refined motion prediction signal for the current block. The video signal is decoded using the refined motion prediction signal as prediction for the current block, wherein a motion vector of the sub-block is generated, and a set of pixel-level motion vector difference values is determined using an affine motion model for the current block.
EFFECT: improved efficiency of video encoding.
11 cl, 42 dwg
