System and method for the dynamic reduction of the entropy of a signal upstream to a data compression device
View Patent ↗A system is proposed for dynamic reduction of the entropy of a signal upstream to a data compression device. The signal includes a set of successive frames. The system has a filtering decision module that provides a setpoint value of filtering and a filtering module that filters the signal according to the setpoint value of filtering and gives a filtered signal to the data compression device. The system further includes a module for obtaining a piece of information on complexity for each frame of the signal. The filtering decision module is adapted to determining the setpoint value of filtering, for each frame of the signal, as a function inter alia of the piece of information on complexity.
1. A system, which dynamically-reduces entropy of a signal upstream to a data compression device, said signal comprising a set of successive frames, said system comprising:
an input receiving the signal comprising the set of successive frames;
an output to the data compression device;
a non-transitory computer-readable medium comprising instructions stored thereon; and
a processing unit configured by the instructions to comprise:
a filtering decision module configured to provide a setpoint value of filtering;
a filtering module receiving the signal from the input and configured to filter the signal according to said setpoint value of filtering and generate a filtered signal on the output to the data compression device; and
a module configured to obtain an estimated complexity K(t) for each frame I(t) of the signal, wherein said filtering decision module is configured to determine said setpoint value of filtering, for each frame of the signal, as a function of said estimated complexity and provide the determined setpoint value to the filtering module,
wherein said estimated complexity K(t) is defined by:
K
(
t
)
=
Q
step
×
1
A
×
∑
k
=
t
t
+
N
∑
T
ε
E
Type
(
P
(
Type
I
(
k
)
=
T
)
×
C
frame
=
T
(
k
)
)
with:
Qstep being a determined quantization step;
A being a standardization factor;
E Type ={inter (P), intra (I), bi-predicted inter (B), bi-predicted and reference inter (Bref)}, being the set of a possible encoding types, i.e. the set of possible values of the variable T;
P(Type I(k)=T) being the probability that a frame I(k) is of a T type;
C frame=T (k) being an estimated cost of a frame I(k) of a T type; and
0≦N≦Nmax, N being the number of next frames I(t+1) to I(t+N), used with the frame I(t) to compute the estimated complexity K(t) of the frame I(t).
2. The system according to claim 1 , wherein the setpoint value of filtering is a setpoint value of resolution and wherein the filtering module is adapted to carrying out a resizing of each frame, according to said setpoint value of resolution.
3. The system according to claim 1 , wherein the setpoint value of filtering is a setpoint value of filtering strength and wherein the filtering module is adapted to carrying out a low-pass filtering of each frame, according to said setpoint value of filtering strength.
4. The system according to claim 1 , wherein said estimated cost C(t) of the frame I(t) is defined by:
C
frame
=
T
=
∑
blk
=
0
Nblock
C
Block
❘
T
(
blk
)
with:
C Bloc|T (blk) being the estimated cost of a block blk included in the frame I(t) and the encoding type T of which is included in the group comprising: “intra”, “inter”, “bi-predicted inter” and “bi-predicted and reference inter”; and
Nbloc is the number of blocks included in the frame I(t).
5. The system according to claim 4 , wherein said estimated cost C Block|T of a block blk is defined by:
C Block|T (blk)= P (blk=tempo)× C tempo (blk)+(1− P (blk=tempo))× C spatio (blk)
with:
C tempo is a cost of encoding with a temporal prediction for the block blk;
C spatio is a cost of encoding with a spatial prediction for the block blk;
P(blk=tempo) is the probability that the block blk is encoded with a temporal prediction.
6. The system according to claim 5 , wherein said cost C tempo is defined by:
C
tempo
(
blk
)
=
α
×
SAD
(
blk
)
Qstep
+
λ
×
MVCost
(
blk
)
with:
α and λ being predetermined parameters;
MV Cost (blk) being a cost of prediction of a motion vector for the block blk;
SAD (blk) being a error of prediction of a motion vector for the block blk.
7. The system according to claim 5 , wherein said cost C spatio is defined by:
C
spatio
(
blk
)
=
β
×
Energy
(
blk
)
Qstep
with:
β being a predetermined parameter;
Energy (blk) being a measurement of spatial energy for the block blk.
8. The system according to claim 6 , wherein said probability P(blk=tempo) is a function of SAD(blk) and the type T of encoding of the frame I(t) in which the block blk is included, and is defined by:
P (blk=tempo|SAD(blk),Frame= I )=0
P (blk=tempo|SAD(blk),Frame= B )=1
Or
P
(
blk
=
tempo
❘
SAD
(
blk
)
,
Frame
=
B
)
=
ⅇ
-
SAD
γ
′
σ
′
P
(
blk
=
tempo
❘
SAD
(
blk
)
,
Frame
=
P
or
Frame
=
Bref
)
=
ⅇ
-
SAD
γ
σ
with:
Frame=1, if the frame I(t) is an “intra” frame;
Frame=P, if the frame I(t) is an “inter” frame;
Frame=B, if the frame I(t) is a “bi-predicted inter” frame;
Frame=Bref, if the frame I(t) is a “bi-predicted and reference inter” frame;
γ′, σ′, γ and σ are predetermined parameters.
9. The system according to claim 1 , wherein the module for obtaining the estimated or computed complexity comprises an encoding module enabling the computation of the complexity associated with a frame I(t) by computing the number of bits really necessary for encoding the frame I(t) and wherein said filtering decision module is adapted to determining said setpoint value of filtering for a frame I(t) as a function of the computed complexity.
10. The system according to claim 1 , wherein the module for obtaining the estimated or computed complexity comprises:
a reading module for reading, in said signal, a size C(t) of a frame I(t) and a quantization step Qstep;
a module for computing the complexity associated with the frame I(t) by multiplying the size read C(t) by the quantization step Qstep.
11. A method comprising:
dynamic reduction of entropy of a signal upstream to a data compression device, said signal comprising a set of successive frames, and said dynamic reduction comprising:
receiving the signal comprising the set of successive frames by a dynamic entropy reduction device;
filtering the signal, by the dynamic entropy reduction device, according to a setpoint value of filtering to produce a filtered signal;
providing the filtered signal from the dynamic entropy reduction device to the data compression device; and
for each frame I(t) of the signal:
obtaining an estimated complexity K(t); and
determining, by the dynamic entropy reduction device, the setpoint value of filtering as a function of the estimated complexity,
wherein said estimated complexity K(t) is defined by:
K
(
t
)
=
Q
step
×
1
A
×
∑
k
=
t
t
+
N
∑
T
∈
E
Type
(
P
(
T
ype
I
(
k
)
=
T
)
×
C
frame
=
T
(
k
)
)
with:
Qstep being a determined quantization step;
A being a standardization factor;
E Type ={inter (P), intra (I), bi-predicted inter (B, bi-predicted and reference inter (Bref)}, being the set of a possible encoding types, i.e. the set of possible values of the variable T;
P(Type I(k)=T) being the probability that a frame I(k) is of a T type;
C frame=T (k) being an estimated cost of a frame I(k) of a T type; and
0≦N≦Nmax, N being the number of next frames I(t+1) to I(t+N), used with the frame I(t) to compute the estimated complexity K(t) of the frame I(t).
12. The method according to claim 11 , wherein the setpoint value of filtering is a setpoint value of resolution and wherein the filtering step comprises a resizing of each frame, according to said setpoint value of resolution.
13. The method according to claim 11 , wherein the setpoint value of filtering is a setpoint value of filtering strength and wherein the filtering step comprises a low-pass filtering of each frame according to said setpoint value of filtering strength.
14. A computer-readable, non-transient storage medium storing a computer program comprising a set of executable instructions executable by a computer or a processor of a dynamic entropy reduction device to implement a method comprising:
dynamic reduction of entropy of a signal upstream to a data compression device, said signal comprising a set of successive frames, and said dynamic reduction comprising:
filtering, by the dynamic entropy reduction device, the signal according to a setpoint value of filtering to produce a filtered signal;
providing the filtered signal from the dynamic entropy reduction device to the data compression device; and
for each frame I(t) of the signal:
obtaining an estimated complexity K(t); and
determining, by the dynamic entropy reduction device, the setpoint value of filtering as a function of the estimated complexity,
wherein said estimated complexity K(t) is defined by:
K
(
t
)
=
Q
step
×
1
A
×
∑
k
=
t
t
+
N
∑
T
∈
E
Type
(
P
(
Type
I
(
k
)
=
T
)
×
C
frame
=
T
(
k
)
)
with:
Qstep being a determined quantization step;
A being a standardization factor;
E Type ={inter (P), intra (I), bi-predicted inter (B), bi-predicted and reference inter (Bref)}, being the set of a possible encoding types, i.e. the set of possible values of the variable T;
P(Type I(k)=T) being the probability that a frame I(k) is of a T type;
C frame=T (k) being an estimated cost of a frame I(k) of a T type; and
0≦N≦Nmax, N being the number of next frames I(t+1) to I(t+N), used with the frame I(t) to compute the estimated complexity K(t) of the frame I(t).