Video coding method of exploiting the temporal redundancy between successive frames
The invention relates to a video coding method of exploiting the temporal redundancy between successive frames in a video sequence. A reference frame, called I-frame, is first approximated by a collection of geometric features, called atoms. The following predicted frames called, P-frames, are approximated by the geometric transformations of the geometric features (atoms) describing the previous frame. Preferably, the I-frame is approximated by a linear combination of N atoms (formula), selected in a redundant, structured library. They are indexed by a string of parameters representing the geometric transformations applied to the generating mother function g(x,y) and the cn are weighting coefficients.
1. Video coding method of exploiting temporal redundancy between successive frames in a video sequence, comprising the steps wherein a reference frame, called an I-frame, is first approximated by a collection of basis functions, called atoms, and wherein either the atoms are quantized, entropy coded, and sent to a decoder or an original I-frame is encoded and transmitted to the decoder using a frame codec, and wherein following predicted frames, called P-frames, are approximated by geometric transformations of the atoms describing a previous frame, wherein the geometric transformations include translations, anisotropic dilations, and rotations, and are applied to a generating mother function g(x,y) by means of the following change of variables:
g
Y
(
x
,
y
)
=
1
a
1
a
2
g
(
x
n
,
y
n
)
,
where
x
n
=
cos
ϑ
(
x
-
b
1
)
-
sin
ϑ
(
y
-
b
2
)
a
1
y
n
=
sin
ϑ
(
x
-
b
1
)
+
cos
ϑ
(
y
-
b
2
)
a
2
and wherein parameters of the geometric transformation are quantized, entropy coded, and sent to the decoder in order to reconstruct the predicted frames.
2. Video coding method according to claim 1 , wherein the collection of atoms is a decomposition of the I-frame obtained using a Matching Pursuit algorithm.
3. Video coding method according to claim 1 , wherein parameters and coefficients of the atoms are quantized and entropy coded.
4. Video coding method according to claim 3 , wherein the quantization of the parameters and the coefficients vary across time, and the variation is controlled by a rate control unit.
5. Video coding method according to claim 1 , further comprising using a residual frame based texture codec that encodes differences between original frames and the frames reconstructed using compensated atoms.
6. Video coding method according to claim 1 , wherein the atoms of the I-frame are computed from quantized frames at an encoder and the decoder and are not transmitted.
7. Video coding method according to claim 1 , wherein the atoms are re-computed after each quantized frame at an encoder and decoder and replace a previous prediction.
8. Video coding method according to claim 1 , wherein the generating mother function is of the following form:
g
(
x
,
y
)
=
(
1
-
x
2
)
exp
(
-
x
2
+
y
2
2
)
.
9. Video coding method according to claim 1 , wherein the I-frame is approximated by a linear combination of N atoms g γn (x,y):
I
(
x
,
y
)
=
∑
n
=
0
N
-
1
c
n
g
γ
n
(
x
,
y
)
,
selected in a redundant, structured library and indexed by a string of parameters γ n representing the geometric transformations applied to the generating mother function g(x,y) Where c n are weighting coefficients.
10. An apparatus for video coding, the apparatus comprising:
a memory containing instructions; and
a processor, operably connected to the memory, that executes the instructions to perform operations comprising:
approximating a reference frame, called an I-frame, using a collection of basis functions, called atoms;
performing one of a group comprising:
quantizing and entropy coding the atoms and transmitting to a decoder, and
encoding an original I-frame and transmitting to the decoder using a frame codec;
approximating following predicted frames, called P-frames, by geometric transformations of the atoms describing a previous frame, wherein the geometric transformations include translations, anisotropic dilations, and rotations; and
applying the predicted frames to a generating mother function g(x,y) by means of the following change of variables:
g
y
(
x
,
y
)
=
1
a
1
a
2
g
(
x
n
,
y
n
)
,
where
x
n
=
cos
ϑ
(
x
-
b
1
)
-
sin
ϑ
(
y
-
b
2
)
a
1
y
n
=
sin
ϑ
(
x
-
b
1
)
-
cos
ϑ
(
y
-
b
2
)
a
2
wherein parameters of the geometric transformation are quantized, entropy coded, and sent to the decoder in order to reconstruct the predicted frames.
11. The apparatus of claim 10 , wherein the collection of atoms is a decomposition of the I-frame obtained using a Matching Pursuit algorithm.
12. The apparatus of claim 10 , wherein parameters and coefficients of the atoms are quantized and entropy coded.
13. The apparatus of claim 12 , wherein thequantization of the parameters and the coefficients vary across time, and the variation is controlled by a rate control unit.
14. The apparatus of claim 10 , the operations further comprising:
using a residual frame based texture codec that encodes differences between original frames and the frames reconstructed using compensated atoms.
15. The apparatus of claim 10 , wherein the atoms of the I-frame are computed from quantized frames at an encoder and the decoder and are not transmitted.
16. The apparatus of claim 10 , wherein the atoms are re-computed after each quantized frame at an encoder and decoder and replace a previous prediction.