Method and device for processing coding unit in image coding system
A method for decoding a picture, according to the present invention, comprises the steps of: receiving split flag information and merge flag information; splitting a coding unit (CU) of one or more depths n into CUs of depth n+1, on the basis of the split flag information; generating a merged CU by combining parts of the CUs of the depth n+1, on the basis of the merge flag information; and performing a decoding procedure for generating a restoration picture, on the basis of the generated merge CU. According to the present invention, a decoding procedure can be performed by merging CUs having similar characteristics, and thus auxiliary information to be received and processed is reduced and coding efficiency is improved.
1. A method for decoding a picture by a decoder, the method comprising:
receiving split flag information, and quad-tree merge flag information;
splitting, based on the split flag information, a coding unit (CU) on a depth n into CUs on a depth n+1 in a quad-tree structure, wherein the CUs on the depth n+1 include a first CU on the depth n+1 and a second CU on the depth n+1;
splitting, based on the split flag information, the second CU on the depth n+1 into CUs on a depth n+2 in the quad-tree structure;
generating, based on the quad-tree merge flag information, a merged CU by merging the first CU on the depth n+1 with at least one of the CUs on the depth n+2, wherein an intra prediction is applied to the merged CU;
generating, based on the intra prediction, a prediction block related to the merged CU; and
generating a reconstructed block based on the prediction block and a residual block,
wherein the at least one of the CUs on the depth n+2 is located in a region of the second CU on the depth n+1, and
wherein a size of the first CU on the depth n+1 is different from a size of each of the CUs on the depth n+2.
2. The method of claim 1 , wherein the first CU on the depth n+1 and the at least one of the CUs on the depth n+2 are adjacent to each other.
3. The method of claim 1 , wherein the merged CU has a square structure that traverses the first CU on the depth n+1 and the second CU on the depth n+1.
4. The method of claim 1 , wherein the merged CU is generated by merging the first CU on a depth n+1 and two of the CUs on the depth n+2.
5. The method of claim 1 , wherein, based on the size of the CUs on the depth n+1 being N×N, a size of the merged CU is 3N×3N.
6. The method of claim 1 , further comprising:
receiving merge type information based on a value of the quad-tree merge flag information indicating 1; and
determining a merge type of the merged CU based on the merge type information.
7. The method of claim 1 , wherein the first CU on the depth n+1 and the at least one of the CUs on the depth n+2 that are merged into the merged CU include pixels for a same object.
8. The method of claim 1 , wherein a value of a split flag for the first CU on the depth n+1 is zero and a value of a split flag for the second CU on the depth n+1 is one.