Affine motion prediction-based image decoding method and device using affine merge candidate list in image coding system
An image decoding method performed by means of a decoding device according to the present disclosure comprises the steps of: configuring an affine merge candidate list with respect to a current block; deriving CPMVs with respect to CPs of the current block on the basis of the affine merge candidate list; deriving prediction samples with respect to the current block on the basis of the CPMVs; and generating a reconstructed picture with respect to the current block on the basis of the derived prediction samples, wherein the maximum number of inherited affine candidates is two, a first inherited affine candidate is derived on the basis of a left block group of the current block, and a second inherited affine candidate is derived on the basis of an upper block group of the current block.
1. An image decoding method, by a decoding apparatus, comprising:
deriving a first inherited affine candidate based on a left block group including a bottom-left corner neighboring block and a left neighboring block of a current block;
deriving a second inherited affine candidate based on a top block group including a top-right corner neighboring block, a top neighboring block and a top-left corner neighboring block of the current block;
constructing a merge candidate list for deriving motion information of subblock units of the current block based on the first inherited affine candidate and the second inherited affine candidate;
deriving control point motion vectors (CPMVs) for control points (CPs) of the current block based on the merge candidate list;
deriving prediction samples for the current block based on the CPMVs; and
generating a reconstructed picture for the current block based on the derived prediction samples,
wherein the CPs include a CP 0 , a CP 1 , and a CP 2 , and
wherein the CP 0 is a point at a top left position of the current block, the CP 1 is a point at a top right position of the current block, and the CP 2 is a point at a bottom left position of the current block.
2. The image decoding method of claim 1 , wherein
neighboring blocks in the left block group are checked in a first order,
the first inherited affine candidate is derived based on a neighboring block coded by a first checked affine motion model, and
the first order is an order from the bottom left corner neighboring block to the left neighboring block.
3. The image decoding method of claim 1 , wherein
neighboring blocks of the top block group is checked in a second order,
the second inherited affine candidate is derived based on a neighboring block coded by a first checked affine motion model, and
the second order is an order from the top right corner neighboring block, to the top neighboring block, and to the bottom left corner neighboring block.
4. The image decoding method of claim 1 , wherein
the merge candidate list includes a constructed affine candidates,
the constructed affine candidates are derived based on the candidate motion vectors for the CPs, and
candidate motion vectors for the CPs are derived based on neighboring blocks of the current block.
5. The image decoding method of claim 4 , wherein
a candidate motion vector for the CP 0 is derived as a motion vector of a first neighboring block determined to be available first when neighboring blocks of a first group are checked in a first order,
a candidate motion vector for the CP 1 is derived as a motion vector of a second neighboring block determined to be available first when neighboring blocks of a second group are checked in a second order, and
a candidate motion vector for the CP 2 is derived as a motion vector of a third neighboring block determined to be available first when neighboring blocks of a third group are checked in a third order.
6. The image decoding method of claim 5 , wherein
the first group includes a neighboring block A, a neighboring block B, and a neighboring block C,
the second group includes a neighboring block D and a neighboring block E, and
the third group includes a neighboring block F and a neighboring block G, and
when a size of the current block is W×H and an x component and a y component of a top left sample position of the current block are 0, the neighboring block A is a block including a sample of coordinates (−1, −1), the neighboring block B is a block including a sample of coordinates (0, −1), the neighboring block C is a block including a sample of coordinates (−1, 0), the neighboring block D is a block including a sample of coordinates (W−1, −1), the neighboring block E is a block including a sample of coordinates (W, −1), the neighboring block F is a block including a sample of coordinates (−1, H−1), and the neighboring block G is a block including a sample of coordinates (−1, H).
7. The image decoding method of claim 6 , wherein
the first order is an order from the neighboring block A to the neighboring block B and to the neighboring block C in the first group,
the second order is an order from the neighboring block D to the neighboring block E in the second group, and
the third order is an order from the neighboring block F to the neighboring block G in the third group.
8. The image decoding method of claim 7 , wherein
the constructed affine candidates comprise a first constructed affine candidate including the candidate motion vector for the CP 1 , the candidate motion vector for the CP 2 , and the candidate motion vector for the CP 3 , when a reference picture for the first neighboring block, a reference picture for the second neighboring block, and a reference picture for the third neighboring block are the same.
9. The image decoding method of claim 7 , wherein
the constructed affine candidates comprise a second constructed affine candidate including the candidate motion vector for the CP 0 and the candidate motion vector for the CP 1 , when the reference picture for the first neighboring block and the reference picture for the second neighboring block are the same.
10. The image decoding method of claim 7 , wherein
the constructed affine candidates comprise a third constructed affine candidate including the candidate motion vector for the CP 0 and the candidate motion vector for the CP 2 , when the reference picture for the first neighboring block and the reference picture for the third neighboring block are the same.
11. The image decoding method of claim 4 , wherein
the merge candidate list comprises a zero candidate when a number of inherited affine candidates and the constructed affine candidates is smaller than a maximum candidate number of the merge candidate list, and the zero candidate comprises candidate motion vectors having a value of 0 and a reference picture index having a value of 0.
12. An image encoding method, by an encoding apparatus, comprising:
deriving a first inherited affine candidate based on a left block group including a bottom-left corner neighboring block and a left neighboring block of a current block;
deriving a second inherited affine candidate based on a top block group including a top-right corner neighboring block, a top neighboring block and a top-left corner neighboring block of the current block;
constructing a merge candidate list for deriving motion information of subblock units of the current block based on the first inherited affine candidate and the second inherited affine candidate;
deriving control point motion vectors (CPMVs) for control points (CPs) of the current block based on the merge candidate list; and
encoding image information including prediction information for the current block,
wherein the CPs include a CP 0 , a CP 1 , and a CP 2 , and
wherein the CP 0 is a point at a top left position of the current block, the CP 1 is a point at a top right position of the current block, and the CP 2 is a point at a bottom left position of the current block.
13. The image encoding method of claim 12 , wherein
neighboring blocks in the left block group are checked in a first order,
the first inherited affine candidate is derived based on a neighboring block coded by a first checked affine motion model, and
the first order is an order from the bottom left corner neighboring block to the left neighboring block.
14. The image encoding method of claim 12 , wherein
neighboring blocks of the top block group is checked in a second order,
the second inherited affine candidate is derived based on a neighboring block coded by a first checked affine motion model, and
the second order is an order from the top right corner neighboring block, to the top neighboring block, and to the bottom left corner neighboring block.
15. A non-transitory computer-readable storage medium storing a bitstream generated by a method, the method comprising:
deriving a first inherited affine candidate based on a left block group including a bottom-left corner neighboring block and a left neighboring block of a current block;
deriving a second inherited affine candidate based on a top block group including a top-right corner neighboring block, a top neighboring block and a top-left corner neighboring block of the current block;
constructing a merge candidate list for deriving motion information of subblock units of the current block based on the first inherited affine candidate and the second inherited affine candidate;
deriving control point motion vectors (CPMVs) for control points (CPs) of the current block based on the merge candidate list;
encoding image information including prediction information for the current block; and
generating the bitstream including the image information,
wherein the CPs include a CP 0 , a CP 1 , and a CP 2 , and
wherein the CP 0 is a point at a top left position of the current block, the CP 1 is a point at a top right position of the current block, and the CP 2 is a point at a bottom left position of the current block.