Method and apparatus for residual sign prediction in transform domain
The present disclosure provides embodiments for coding and decoding signs of transformation coefficients which is applicable, for instance, in image and/or video coding and decoding. In particular, a plurality of the signs are predicted and only a prediction error signal is embedded in the bitstream. The prediction error signal may have a distribution which can be efficiently coded with CABAC or another variable length (entropy) coding. In order to perform the prediction of the signs efficiently, the signs of a plurality of coefficients of a transformed image block are predicted based on a cost function including transformed difference between adjacent pixels neighboring the transformed image block and prediction of the adjacent pixels calculated based on a prediction signal of the image block.
1. An apparatus for decoding an image block comprising processing circuitry, wherein the processing circuitry is configured for:
predicting signs of a plurality of non-zero coefficients of a transformed image block based on a cost function, wherein the cost function includes a transformed difference between adjacent pixels neighboring the transformed image block and a prediction of the adjacent pixels calculated based on a prediction signal of the image block, wherein the adjacent pixels include pixels in two rows and two columns in adjacent blocks closest to a boundary with the image block, wherein the transformed difference in a transform domain is obtained by a one-dimensional orthogonal transform, and wherein the cost function is calculated in the transform domain for all possible sets of combinations of the signs and a set minimizing the cost function is selected as the predicted signs; and
reconstructing the signs of the plurality of coefficients according to the predicted signs,
wherein the transformed difference between the adjacent pixels neighboring the transformed image block comprises a sum of squared transformed differences between the adjacent pixels neighboring the transformed image block, and
wherein the predicting the signs of the plurality of coefficients of the transformed image block comprises:
calculating the transformed difference between the adjacent pixels neighboring the transformed image block;
calculating the prediction of the adjacent pixels calculated based on the prediction signal of the image block;
calculating the cost function, wherein the transformed image block is reconstructed according to each set of signs out of the sets of signs; and
selecting as the predicted signs the set of the signs that minimizes a cost given by the cost function.
2. The apparatus for decoding the image block according to claim 1 , wherein the processing circuitry is further configured for:
parsing, from a coded stream, a sign prediction error; and
before the reconstructing the signs of the plurality of coefficients, adding the sign prediction error to the predicted signs.
3. The apparatus for decoding the image block according to claim 1 , wherein the transformed image block is reconstructed by adding the set of the signs to coefficients parsed from a coded stream.
4. The apparatus for decoding the image block according to claim 1 , wherein the cost function (F) is given by:
F
=
∑
n
=
0
N
-
1
(
t
n
-
q
n
)
2
+
∑
m
=
0
M
-
1
(
v
m
-
o
m
)
2
.
with t n =Trans1D(T n ), q n =Trans1D(Q n ), v m =Trans1D(V m ), and o m =Trans1D(O m ), wherein Trans1D(.) is a one-dimensional orthogonal transform and T n =[2X n,-1 −X n,-2 −P n,0 ], V m =[2Z −1,m −Z −2,m −P 0,m ], Q n =R n,0 and O m =R 0,m , wherein P is prediction signal, and X and Z are neighboring pixels, and N and M being height and width of the block in which the signs are predicted, respectively.
5. The apparatus for decoding the image block according to claim 1 , wherein the adjacent pixels are located at a horizontal and a vertical boundary of the image block.
6. The apparatus for decoding the image block according to claim 2 , wherein the processing circuitry is further configured for:
decoding the sign prediction error using Context-Adaptive Binary Arithmetic Coding (CABAC).
7. An apparatus for encoding an image block comprising processing circuitry, wherein the processing circuitry is configured for:
predicting signs of a plurality of non-zero coefficients of a transformed image block based on a cost function, wherein the cost function comprises a transformed difference between adjacent pixels neighboring the transformed image block and a prediction of the adjacent pixels calculated based on a prediction signal of the image block, wherein the adjacent pixels include pixels in two rows and two columns in adjacent blocks closest to a boundary with the image block, wherein the transformed difference in a transform domain is obtained by a one-dimensional orthogonal transform, and wherein the cost function is calculated in the transform domain for all possible sets of combinations of the signs and a set minimizing the cost function is selected as the predicted signs; and
encoding the signs of the plurality of coefficients according to the predicted signs,
wherein the transformed difference between the adjacent pixels neighboring the transformed image block is a sum of squared transformed differences between the adjacent pixels neighboring the transformed image block, and
wherein the predicting the signs of the plurality of coefficients of the transformed image block comprises:
calculating the transformed difference between the adjacent pixels neighboring the transformed image block;
calculating the prediction of the adjacent pixels calculated based on the prediction signal of the image block;
calculating the cost function, wherein the transformed image block is reconstructed according to each set of signs out of the sets of signs; and
selecting as the predicted signs that set of the signs which minimize the cost given by the cost function.
8. The apparatus for encoding the image block according to claim 7 , wherein the processing circuitry is specifically configured for encoding the signs of the plurality of coefficients by:
determining a sign prediction error as a difference between the signs of the plurality of coefficients and the predicted signs of the plurality of coefficients; and
inserting the sign prediction error into a coded stream including the coded image block.
9. The apparatus for encoding the image block according to claim 7 , wherein the cost function (F) is given by:
F
=
∑
n
=
0
N
-
1
(
t
n
-
q
n
)
2
+
∑
m
=
0
M
-
1
(
v
m
-
o
m
)
2
.
with t n =Trans1D(T n ), q n =Trans1D(Q n ), v m =Trans1D(V m ), and o m =Trans1D(O m ), wherein Trans1D(.) is a one-dimensional orthogonal transform and T n =[2X n,-1 −X n,-2 −P n,0 ], V m =[2Z −1,m −Z −2,m −P 0,m ], Q n =R n,0 and O m =R 0,m , wherein P is prediction signal, and X and Z are neighboring pixels, and N and M being height and width of the block in which the signs are predicted, respectively.
10. The apparatus for encoding the image block according to claim 7 , wherein the adjacent pixels are located at a horizontal and a vertical boundary of the image block.
11. The apparatus for encoding the image block according to claim 8 , wherein the processing circuitry is further configured for encoding the sign prediction error using Context-Adaptive Binary Arithmetic Coding (CABAC).
12. A method for decoding an image block, the method comprising:
predicting signs of a plurality of non-zero coefficients of a transformed image block based on a cost function, wherein the cost function comprises a transformed difference between adjacent pixels neighboring the transformed image block and a prediction of the adjacent pixels calculated based on a prediction signal of the image block, wherein the adjacent pixels include pixels in two rows and two columns in adjacent blocks closest to a boundary with the image block, wherein the transformed difference in a transform domain is obtained by a one-dimensional orthogonal transform, and wherein the cost function is calculated in the transform domain for all possible sets of combinations of the signs and a set minimizing the cost function is selected as the predicted signs; and
reconstructing the signs of the plurality of coefficients according to the predicted signs,
wherein the transformed difference between the adjacent pixels neighboring the transformed image block comprises a sum of squared transformed differences between the adjacent pixels neighboring the transformed image block, and
wherein the predicting the signs of the plurality of coefficients of the transformed image block comprises:
calculating the transformed difference between the adjacent pixels neighboring the transformed image block;
calculating the prediction of the adjacent pixels calculated based on the prediction signal of the image block;
calculating the cost function, wherein the transformed image block is reconstructed according to each set of signs out of the sets of signs; and
selecting as the predicted signs the set of the signs that minimizes a cost given by the cost function.
13. A non-transitory computer readable medium storing a program including instructions that, when executed on a processor, configure the processor to perform the method of claim 12 .
14. A method for encoding an image, the method comprising:
predicting signs of a plurality of non-zero coefficients of a transformed image block based on a cost function, wherein the cost function comprises a transformed difference between adjacent pixels neighboring the transformed image block and a prediction of the adjacent pixels calculated based on a prediction signal of the image block, wherein the adjacent pixels including pixels in two rows and two columns in adjacent blocks closest to a boundary with the image block, wherein the transformed difference in a transform domain is obtained by a one-dimensional orthogonal transform, and wherein the cost function is calculated in the transform domain for all possible sets of combinations of the signs and a set minimizing the cost function is selected as the predicted signs; and
encoding the signs of the plurality of coefficients according to the predicted signs,
wherein the transformed difference between the adjacent pixels neighboring the transformed image block is a sum of squared transformed differences between the adjacent pixels neighboring the transformed image block, and
wherein the predicting the signs of the plurality of coefficients of the transformed image block comprises:
calculating the transformed difference between the adjacent pixels neighboring the transformed image block;
calculating the prediction of the adjacent pixels calculated based on the prediction signal of the image block;
calculating the cost function, wherein the transformed image block is reconstructed according to each set of signs out of the sets of signs; and
selecting as the predicted signs that set of the signs which minimize the cost given by the cost function.
15. A non-transitory computer readable medium storing a program including instructions that, when executed on a processor, configure the processor to perform the method claim 14 .