Image decoding method and apparatus based on inter prediction in image coding system
An image decoding method that is performed by a decoding apparatus according to the present disclosure comprises the steps of: forming a merge candidate list based on neighbouring blocks of a current block; deriving costs with respect to merge candidates included in the merge candidate list; deriving a revised merge candidate list based on the costs with respect to the merge candidates; deriving movement information of the current block based on the revised merge candidate list; and performing prediction on the current block based on the movement information, wherein the neighbouring blocks include spatial neighbouring blocks and temporal neighbouring blocks.
1. An image decoding method performed by a decoding device, the method comprising:
constructing a merge candidate list based on a neighboring block of a current block;
deriving costs for merge candidates included in the merge candidate list;
deriving a modified merge candidate list based on the costs for the merge candidates;
deriving motion information of the current block based on the modified merge candidate list; and
performing prediction of the current block based on the motion information,
wherein the neighboring block includes a spatial neighboring block and a temporal neighboring block, and
wherein the deriving of the modified merge candidate list based on the costs for the merge candidates includes:
deriving a reordered merge candidate list by reordering the merge candidates in an order from small to large costs thereof;
deriving a refine merge candidate based on a predetermined merge candidate among the merge candidates; and
deriving the modified merge candidate list by adding the refine merge candidate to the reordered merge candidate list.
2. The image decoding method of claim 1 , wherein a cost of a merge candidate is derived as the sum of absolute differences (SAD) between a template of the current block and a template of the merge candidate.
3. The image decoding method of claim 2 , wherein the template of the current block includes a top template and a left template,
the cost is derived as the sum of a top cost and a left cost,
the top cost is derived as the SAD between the top template of the current block and a top template of the merge candidate, and
the left cost is derived as the SAD between the left template of the current block and a left template of the merge candidate.
4. The image decoding method of claim 3 , wherein size information on the template of the current block is received,
based on a case that a size of the current block is W×H, and an x component of a top-left sample position of the current block is 0 and a y component thereof is 0, the template of the current block includes a top-left sample at coordinates (−n, 0), includes the left template having a size of n×H and a top-left sample at coordinates (0, −n), and is derived as the top template having a size of W×n, and
the n is a value which the size information represents.
5. The image decoding method of claim 3 , wherein the template of the current block is variable according to a neighboring block for the merge candidate.
6. The image decoding method of claim 5 , wherein the spatial neighboring block includes a left neighboring block and a top neighboring block of the current block,
based on a case that a size of the current block is W×H, and an x component of a top-left sample position of the current block is 0 and a y component thereof is 0, the template of the current block for a merge candidate representing motion information of the top neighboring block includes a top-left sample at coordinates (−n, 0), includes the left template having a size of n×H and a top-left sample at (−n, −n), and is derived as the top template having a size of (n+W−1)×n, and
the template of the current block for a merge candidate representing motion information of the left neighboring block includes a top-left sample at coordinates (−n, −n), includes the left template having a size of n×(n+H−1) and a top-left sample at coordinates (0, −n), and is derived as the top template having a size of W×n.
7. The image decoding method of claim 1 , wherein a cost of a merge candidate is derived as the sum of absolute differences (SAD) between a first reference block of the merge candidate and a second reference block of the merge candidate, and
the first reference block is a reference block which a first motion information using a first reference picture list of the merge candidate is related to, and the second reference block is a reference block which a second motion information using a second reference picture list of the merge candidate is related to.
8. The image decoding method of claim 7 , wherein based on a case that the merge candidate includes the second motion information as uni-prediction motion information, motion information that is related to a reference block among reference blocks within a search range, which has a minimum cost with the second reference block which the seconder motion information is related to, is derived as the first motion information.
9. The image decoding method of claim 8 , wherein the search range is a 1-integer-pel range around a position which the second motion information is related to.
10. The image decoding method of claim 1 , wherein the deriving of the refine merge candidate based on the predetermined merge candidate among the merge candidates includes:
deriving a template among templates of reference blocks within a search range, which has a smallest cost with the template of the current block; and
deriving motion information, which is related to a reference block of the derived template, as the refine merge candidate,
wherein the search range is a 1-integer-pel range around a position which motion information represented by the predetermined merge candidate is related to.
11. The image decoding method of claim 1 , wherein the deriving of the refine merge candidate based on the predetermined merge candidate among the merge candidates includes:
deriving, as a first refine reference block and a second refine reference block, a first reference block and a second reference block among first reference blocks within a first search range and second reference blocks within a second search range, which have the smallest sum of absolute differences (SAD); and
deriving, as refine merge candidates, a first motion information related to the first refine reference block and a second motion information related to the second refine reference block, and
wherein the first search range is a 1-integer-pel range around a position which a first motion information using a first reference picture list of the predetermined merge candidate is related to, and
wherein the second search range is a 1-integer-pel range around a position which a second motion information using a second reference picture list of the predetermined merge candidate is related to.
12. The image decoding method of claim 1 , wherein the refine merge candidate is derived as a merge candidate including a location illumination compensation (LIC) flag which has an opposite value to that of an LIC flag of the predetermined merge candidate, and
the LIC flag represents whether the LIC is applied.
13. The image decoding method of claim 1 , further comprising:
determining whether a reordered merge candidate of the modified merge candidate list represents the same motion information as a merge candidate that comes in order before it; and
removing the reordered merge candidate from the modified merge candidate list based on a case that the reordered merge candidate represents the same motion information as the merge candidate that comes in order before it.
14. A decoding device which performs image decoding, the device comprising:
an entropy decoder which obtains information on inter-prediction of a current block through a bitstream; and
a predictor which constructs a merge candidate list based on a neighboring block of the current block, derives costs for merge candidates included in the merge candidate list, derives a modified merge candidate list based on the costs for the merge candidates, derives motion information of the current block based on the modified merge candidate list and a merge index, and performs prediction of the current block based on the motion information,
wherein the neighboring block includes a spatial neighboring block and a temporal neighboring block, and the information on inter-prediction includes the merge index, and
wherein the predictor derives a reordered merge candidate list by reordering the merge candidates in an order from small to large costs thereof, derives a refine merge candidate based on a predetermined merge candidate among the merge candidates, and derives the modified merge candidate list by adding the refine merge candidate to the reordered merge candidate list.
15. An encoding device which performs image encoding, the device comprising:
a predictor which constructs a merge candidate list based on a neighboring block of a current block, derives costs for merge candidates included in the merge candidate list, derives a modified merge candidate list based on the costs for the merge candidates, determines motion information of the current block based on the modified merge candidate list, and performs prediction of the current block based on the motion information; and
an entropy encoder which encodes information on inter-prediction of the current block,
wherein the neighboring block includes a spatial neighboring block and a temporal neighboring block, and the information on inter-prediction includes the merge index, and wherein the predictor derives a reordered merge candidate list by reordering the merge candidates in an order from small to large costs thereof, derives a refine merge candidate based on a predetermined merge candidate among the merge candidates, and derives the modified merge candidate list by adding the refine merge candidate to the reordered merge candidate list.