The present disclosure relates to motion vector refinement. As a first step, an initial motion vector and a template for the block are obtained. Then, the refinement of the initial motion vector is determined by template matching with said template in a search space. The search space is located on a position given by the initial motion vector and includes one or more fractional sample positions, wherein each of the fractional sample positions belonging to the search space is obtained by interpolation filtering with a filter of a predefined tap-size assessing integer samples only within a window, said window being formed by integer samples accessible for the template matching in said search space.

An image decoding method according to the present invention comprises: a step of acquiring an initial motion vector of a current block; a step of deriving a refined motion vector for each of a plurality of search points on the basis of the initial motion vector; and a step of acquiring a motion vector of the current block on the basis of the refined motion vector of any one of the plurality of search points.

An encoder includes circuitry and memory. Using the memory, the circuitry, in inter prediction processing: derives a first motion vector of a current block to be processed, using a motion vector of a previous block which has been previously processed; derives a second motion vector of the current block by performing motion estimation in the vicinity of the first motion vector; and generates a prediction image of the current block by performing motion compensation using the second motion vector.

The present invention relates to the performance of motion vector refinement in a search space for multi-reference inter-prediction. Two or more reference pictures are selected, one of those used for motion vector refinement. Based on an initial estimate of a motion vector to the reference picture for motion vector refinement, a search space in this reference image is constructed. Using template matching, the first motion vector is refined. The second motion vector to another reference picture is calculated using its initial estimate, the initial estimate of the first motion vector and the refined first motion vector.

The present invention relates to the performance of motion vector refinement in a search space for multi-reference inter-prediction. Two or more reference pictures are selected, one of those used for motion vector refinement. Based on an initial estimate of a motion vector to the reference picture for motion vector refinement, a search space in this reference image is constructed. Using template matching, the first motion vector is refined. The second motion vector to another reference picture is calculated using its initial estimate, the initial estimate of the first motion vector and the refined first motion vector.

An image decoding device includes a prediction unit configured to generate a prediction signal included in a prediction block based on a motion vector. The prediction unit is configured to perform refinement processing of setting a search range based on a reference position specified by the motion vector, specifying a corrected reference position having the smallest predetermined cost from the search range, and correcting the motion vector based on the corrected reference position. When a block size of the prediction block is larger than a predetermined block size, the prediction unit is configured to divide the prediction block into sub-block groups and perform the refinement processing for each sub-block.

The present invention relates to the performance of motion vector refinement in a search space for multi-reference inter-prediction. Two or more reference pictures are selected, one of those used for motion vector refinement. Based on an initial estimate of a motion vector to the reference picture for motion vector refinement, a search space in this reference image is constructed. Using template matching, the first motion vector is refined. The second motion vector to another reference picture is calculated using its initial estimate, the initial estimate of the first motion vector and the refined first motion vector.

The present invention relates to the performance of motion vector refinement in a search space for multi-reference inter-prediction. Two or more reference pictures are selected, one of those used for motion vector refinement. Based on an initial estimate of a motion vector to the reference picture for motion vector refinement, a search space in this reference image is constructed. Using template matching, the first motion vector is refined. The second motion vector to another reference picture is calculated using its initial estimate, the initial estimate of the first motion vector and the refined first motion vector.

An encoder includes circuitry and memory. Using the memory, the circuitry, in inter prediction processing: derives a first motion vector of a current block to be processed, using a motion vector of a previous block which has been previously processed; derives a second motion vector of the current block by performing motion estimation in the vicinity of the first motion vector; and generates a prediction image of the current block by performing motion compensation using the second motion vector.

An image decoding device includes a prediction unit configured to generate a prediction signal included in a prediction block based on a motion vector. The prediction unit is configured to perform refinement processing of setting a search range based on a reference position specified by the motion vector, specifying a corrected reference position having the smallest predetermined cost from the search range, and correcting the motion vector based on the corrected reference position. When a block size of the prediction block is larger than a predetermined block size, the prediction unit is configured to divide the prediction block into sub-block groups and perform the refinement processing for each sub-block.