Predicting filter coefficients from fixed filters for video coding
An example device for filtering a decoded block of video data includes one or more processing units configured to construct a plurality of filters for classes of blocks of a current picture of video data. To construct the plurality of filters for each of the classes, the processing units are configured to determine a value of a flag that indicates whether a fixed filter is used to predict a set of filter coefficients of the class, and in response to the fixed filter being used to predict the set of filter coefficients, determine an index value into a set of fixed filters and predict the set of filter coefficients of the class using a fixed filter of the set of fixed filters identified by the index value.
1 . A method of filtering a decoded block of video data, the method comprising:
decoding a current block of a current picture of the video data;
selecting a filter to be used to filter at least one pixel of the current block wherein f(k, l) represents the selected filter, and R(i, j) represents a pixel at position (i, j) of the current picture;
selecting a geometric transform to be performed on one of a filter support region or coefficients of the selected filter, wherein the geometric transform comprises one of a rotation transform, a diagonal flip transform, or a vertical flip, and wherein selecting the geometric transform comprises:
calculating a horizontal gradient g h according to:
g
h
=
∑
k
=
i
-
M
i
+
N
+
M
-
1
∑
l
=
j
-
M
j
+
N
+
M
-
1
H
k
,
l
,
where
H
k
,
l
=
❘
"\[LeftBracketingBar]"
2
R
(
k
,
l
)
-
R
(
k
-
1
,
l
)
-
R
(
k
+
1
,
l
)
❘
"\[RightBracketingBar]"
;
calculating a vertical gradient g v according to:
g
v
=
∑
k
=
i
-
M
i
+
N
+
M
-
1
∑
l
=
j
-
M
j
+
N
+
M
-
1
V
k
,
l
,
where
V
k
,
l
=
❘
"\[LeftBracketingBar]"
2
R
(
k
,
l
)
-
R
(
k
,
l
-
1
)
-
R
(
k
,
l
+
1
)
❘
"\[RightBracketingBar]"
;
calculating a first diagonal gradient g d1 according to:
g
d
1
=
∑
k
=
i
-
M
i
+
N
+
M
-
1
∑
l
=
j
-
M
j
+
N
+
M
-
1
D
1
k
,
l
,
where
D
1
k
,
l
=
❘
"\[LeftBracketingBar]"
2
R
(
k
,
l
)
-
R
(
k
-
1
,
l
-
1
)
-
R
(
k
+
1
,
l
+
1
)
❘
"\[RightBracketingBar]"
;
calculating a second diagonal gradient g d2 according to:
g
d
2
=
∑
k
=
i
-
M
i
+
N
+
M
-
1
∑
j
=
j
-
M
j
+
N
+
M
-
1
D
2
k
,
l
,
where
D
2
k
,
l
=
❘
"\[LeftBracketingBar]"
2
R
(
k
,
l
)
-
R
(
k
-
1
,
l
+
1
)
-
R
(
k
+
1
,
l
-
1
)
❘
"\[RightBracketingBar]"
;
and
selecting the diagonal flip transform when g d2 <g d1 and g v <g h ;
selecting the vertical flip transform when g d1 <g d2 and g h <g v ; or
selecting the rotation transform when g d1 <g d2 and g v <g h ;
performing the geometric transform on either the filter support region or the coefficients of the selected filter; and
filtering the at least one pixel of the current block using the selected filter after performing the geometric transform.
2 . The method of claim 1 ,
wherein the rotation transform comprises f R (k, l)=f(K−l−1,k),
wherein the diagonal flip transform comprises f D (k, l)=f(l, k),
wherein the vertical flip transform comprises f V (k, l)=f(k, K−l−1), and
wherein K is the size of the selected filter, k and l are coordinates of coefficients of the selected filter or coordinates of values in the filter support region, 0≤k, 1≤K−1, location (0, 0) is at the upper left corner of the selected filter or the filter support region, and location (K−1, K−1) is at the lower right corner of the upper left corner of the selected filter or the filter support region.
3 . The method of claim 1 , wherein the filter support region comprises a plurality of neighboring pixels to the at least one pixel of the current block to which coefficients of the selected filter are to be applied, and filtering the at least one pixel comprises performing the geometric transform on either the filter support region or the coefficients of the selected filter.
4 . The method of claim 1 , further comprising encoding the current block prior to decoding the current block.
5 . A device for filtering a decoded block of video data, the device comprising:
a memory configured to store the video data; and
a processing implemented in circuitry and configured to:
decode a current block of a current picture of the video data;
select a filter to be used to filter at least one pixel of the current block, wherein f(k, l) represents the selected filter, and R(i, j) represents a pixel at position (i, j) of the current picture;
select a geometric transform to be performed on one of a filter support region or coefficients of the selected filter, wherein the geometric transform comprises one of a rotation transform, a diagonal flip transform, or a vertical flip, and to select the geometric transform, the processing system is configured to:
calculate a horizontal gradient g h according to:
g
h
=
∑
k
=
i
-
M
i
+
N
+
M
-
1
∑
l
=
j
-
M
j
+
N
+
M
-
1
H
k
,
l
,
where
H
k
,
l
=
❘
"\[LeftBracketingBar]"
2
R
(
k
,
l
)
-
R
(
k
-
1
,
l
)
-
R
(
k
+
1
,
l
)
❘
"\[RightBracketingBar]"
;
calculate a vertical gradient g v according to:
g
v
=
∑
k
=
i
-
M
i
+
N
+
M
-
1
∑
l
=
j
-
M
j
+
N
+
M
-
1
V
k
,
l
,
where
V
k
,
l
=
❘
"\[LeftBracketingBar]"
2
R
(
k
,
l
)
-
R
(
k
,
l
-
1
)
-
R
(
k
,
l
+
1
)
❘
"\[RightBracketingBar]"
;
calculate a first diagonal gradient g d1 according to:
g
d
1
=
∑
k
=
i
-
M
i
+
N
+
M
-
1
∑
l
=
j
-
M
j
+
N
+
M
-
1
D
1
k
,
l
,
where
D
1
k
,
l
=
❘
"\[LeftBracketingBar]"
2
R
(
k
,
l
)
-
R
(
k
-
1
,
l
-
1
)
-
R
(
k
+
1
,
l
+
1
)
❘
"\[RightBracketingBar]"
;
calculate a second diagonal gradient g d2 according to:
g
d
2
=
∑
k
=
i
-
M
i
+
N
+
M
-
1
∑
j
=
j
-
M
j
+
N
+
M
-
1
D
2
k
,
l
,
where
D
2
k
,
l
=
❘
"\[LeftBracketingBar]"
2
R
(
k
,
l
)
-
R
(
k
-
1
,
l
+
1
)
-
R
(
k
+
1
,
l
-
1
)
❘
"\[RightBracketingBar]"
;
and
select the diagonal flip transform when g d2 <g d1 and g v <g h ;
select the vertical flip transform when g d1 <g d2 and g h <g v ; or
select the rotation transform when g d1 <g d2 and g v <g h ;
perform the geometric transform on either the filter support region or the coefficients of the selected filter; and
filter the at least one pixel of the current block using the selected filter after performing the geometric transform.
6 . The device of claim 5 , wherein the device is a wireless communication device, further comprising:
a receiver configured to receive video data including the current picture.
7 . The device of claim 6 , wherein the wireless communication device is a cellular telephone and the video data is received by the receiver and modulated according to a cellular communication standard.
8 . The device of claim 5 ,
wherein the rotation transform comprises f R (k, l)=f(K−l−1, k),
wherein the diagonal flip transform comprises f D (k, l)=f(l, k),
wherein the vertical flip transform comprises f V (k, l)=f(k, K−l−1), and
wherein K is the size of the selected filter, k and l are coordinates of coefficients of the selected filter or coordinates of values in the filter support region, 0≤k, 1≤K−1, location (0, 0) is at the upper left corner of the selected filter or the filter support region, and location (K−1, K−1) is at the lower right corner of the upper left corner of the selected filter or the filter support region.
9 . The device of claim 5 , wherein the filter support region comprises a plurality of neighboring pixels to the at least one pixel of the current block to which coefficients of the selected filter are to be applied, and to filter the at least one pixel, the processing system is configured to perform the geometric transform on either the filter support region or the coefficients of the selected filter.
10 . The device of claim 5 , wherein the processing system is further configured to encode the current block prior to decoding the current block.
11 . A computer-readable storage medium having stored thereon instructions that, when executed, cause a processor to:
decode a current block of a current picture of the video data;
select a filter to be used to filter at least one pixel of the current block, wherein f(k, l) represents the selected filter, and R(i, j) represents a pixel at position (i, j) of the current picture;
select a geometric transform to be performed on one of a filter support region or coefficients of the selected filter, wherein the geometric transform comprises one of a rotation transform, a diagonal flip transform, or a vertical flip, and to select the geometric transform, the processing system is configured to:
calculate a horizontal gradient g h according to:
g
h
=
∑
k
=
i
-
M
i
+
N
+
M
-
1
∑
l
=
j
-
M
j
+
N
+
M
-
1
H
k
,
l
,
where
H
k
,
l
=
❘
"\[LeftBracketingBar]"
2
R
(
k
,
l
)
-
R
(
k
-
1
,
l
)
-
R
(
k
+
1
,
l
)
❘
"\[RightBracketingBar]"
;
calculate a vertical gradient g v according to:
g
v
=
∑
k
=
i
-
M
i
+
N
+
M
-
1
∑
l
=
j
-
M
j
+
N
+
M
-
1
V
k
,
l
,
where
V
k
,
l
=
❘
"\[LeftBracketingBar]"
2
R
(
k
,
l
)
-
R
(
k
,
l
-
1
)
-
R
(
k
,
l
+
1
)
❘
"\[RightBracketingBar]"
;
calculate a first diagonal gradient g d1 according to:
g
d
1
=
∑
k
=
i
-
M
i
+
N
+
M
-
1
∑
l
=
j
-
M
j
+
N
+
M
-
1
D
1
k
,
l
,
where
D
1
k
,
l
=
❘
"\[LeftBracketingBar]"
2
R
(
k
,
l
)
-
R
(
k
-
1
,
l
-
1
)
-
R
(
k
+
1
,
l
+
1
)
❘
"\[RightBracketingBar]"
;
calculate a second diagonal gradient g d2 according to:
g
d
2
=
∑
k
=
i
-
M
i
+
N
+
M
-
1
∑
j
=
j
-
M
j
+
N
+
M
-
1
D
2
k
,
l
,
where
D
2
k
,
l
=
❘
"\[LeftBracketingBar]"
2
R
(
k
,
l
)
-
R
(
k
-
1
,
l
+
1
)
-
R
(
k
+
1
,
l
-
1
)
❘
"\[RightBracketingBar]"
;
and
select the diagonal flip transform when g d2 <g d1 and g v <g h ;
select the vertical flip transform when g d1 <g d2 and g h <g v ; or
select the rotation transform when g d1 <g d2 and g v <g h ;
perform the geometric transform on either the filter support region or the coefficients of the selected filter; and
filter the at least one pixel of the current block using the selected filter after performing the geometric transform.
12 . The computer-readable storage medium of claim 11 ,
wherein the rotation transform comprises f R (k, l)=f(K−l−1, k),
wherein the diagonal flip transform comprises f D (k, l)=f(l, k),
wherein the vertical flip transform comprises f V (k, l)=f(k, K−l−1), and
wherein K is the size of the selected filter, k and l are coordinates of coefficients of the selected filter or coordinates of values in the filter support region, 0≤k, 1≤K−1, location (0, 0) is at the upper left corner of the selected filter or the filter support region, and location (K−1, K−1) is at the lower right corner of the upper left corner of the selected filter or the filter support region.
13 . The computer-readable storage medium of claim 11 , wherein the filter support region comprises a plurality of neighboring pixels to the at least one pixel of the current block to which coefficients of the selected filter are to be applied, and wherein the instructions that cause the processor to filter the at least one pixel comprise instructions that cause the processor to perform the geometric transform on either the filter support region or the coefficients of the selected filter.
14 . The computer-readable storage medium of claim 11 , further comprising instructions that cause the processor to encode the current block prior to decoding the current block.