Probability update method for binary arithmetic coding/decoding, and entropy coder/decoder using the same
A probably update method used in context-based adaptive binary arithmetic coding (CABAC) is disclosed. The probably update method for binary arithmetic decoding obtains an autocorrelation value of each bin by using values of received bins, determines at least one scaling factor used to update a probability of a binary value based on the autocorrelation value, and updates a probability used in context-based adaptive binary arithmetic decoding by using the determined at least one scaling factor.
1. A probability update method for binary arithmetic decoding, the method comprising:
receiving a predetermined number of bins that are to be binary arithmetic decoded;
obtaining an autocorrelation value of each of the bins by using values of the received predetermined number of bins;
determining at least one scaling factor used to update a probability of a binary value based on the autocorrelation value;
updating a probability used in context-based adaptive binary arithmetic decoding by using the determined at least one scaling factor; and
arithmetic decoding a current bin by using the updated probability,
wherein the updating a probability comprises:
obtaining the updated probability by adding a first value and a second value, the first value being obtained by multiplying a value based on the at least one scaling factor by a probability of a previous bin, and the second value being obtained by multiplying the at least one scaling factor by a value of a current bin,
wherein when a value of the current bin is y (y is 0 or 1), a probability previous to the current bin is p(t−1) (t is an integer), the updated probability is p(t), and the at least one scaling factor is α, the updated probability p(t) is obtained according to an equation P(t)=αy+(1−α)*P(t−1).
2. The probability update method of claim 1 , wherein the at least one scaling factor is determined as a value having a minimum mean square error between a probability of each bin determined based on the autocorrelation value and a value of each bin.
3. The probability update method of claim 1 , wherein, when the autocorrelation value between the bins spaced by a predetermined distance k (k is an integer) is denoted by R k , an average value of the bins is denoted by M (M is a real number), a variance of the bins is denoted by σ, the number of bins is denoted by (N+1) (N is an integer), and values of (N+1) bins are denoted by y j (j is an integer from 0 to N), R k is obtained according to an equation
R
k
=
1
σ
2
∑
j
=
0
N
(
y
j
-
M
)
(
y
j
-
k
-
M
)
.
4. The probability update method of claim 3 , wherein the scaling factor is one, and the one scaling factor α is obtained according to an equation
α
=
(
3
R
k
-
1
)
2
R
k
according to the obtained autocorrelation value R k .
5. The probability update method of claim 1 , wherein the scaling factors are two, and based on a value of the autocorrelation value R k ,
when R k ε[−1, 1/7], the two scaling factors α1 and α2 have a value of 0,
when
R
k
∈
[
1
/
7
,
1
/
2
]
,
α
1
=
3
-
9
-
7
R
k
-
1
R
k
,
α
2
=
0
;
when
R
k
∈
[
1
/
2
,
5
/
7
]
,
α
1
=
1
,
α
2
=
0
;
and
when
R
k
∈
[
5
/
7
,
1
]
,
α
1
=
1
,
α
2
=
3
R
k
-
2
-
2
R
k
2
-
1
R
k
-
1
.
6. The probability update method of claim 1 , wherein when the at least one scaling factor is plural, the plurality of scaling factors are denoted by α i , a value of the current bin is y (y is 0 or 1), a probability previous to the current bin is p(t−1) (t is an integer), and probabilities p i (t) updated according to the scaling factors α i are obtained according to an equation P i (t)=α i y+(1−α i )*P i (t−1), a weight average value of the plurality of updated probabilities p i (t) is used as a final update probability P(t).
7. A probability update method for binary arithmetic decoding, the method comprising:
receiving a predetermined number of bins that are to be binary arithmetic decoded;
obtaining entropy values indicating an average bit value of the bins by using a plurality of probability models having different scaling factors;
determining a scaling factor of a probability model used to obtain a minimum entropy value among the plurality of probability models;
updating a probability used in context-based adaptive binary arithmetic decoding by using the determined scaling factor; and
performing context-based adaptive binary arithmetic decoding using the updated probability,
wherein the updating a probability comprises:
obtaining the updated probability by adding a first value and a second value, the first value being obtained by multiplying a value based on the at least one scaling factor by a probability of a previous bin, and the second value being obtained by multiplying the at least one scaling factor by a value of a current bin,
wherein when a value of the current bin is y (y is 0 or 1), a probability previous to the current bin is p(t−1) (t is an integer), the updated probability is p(t), and the at least one scaling factor is α, the updated probability p(t) is obtained according to an equation P(t)=αy+(1−α)*P(t−1).
8. The probability update method of claim 7 , wherein when a value of a current bin is y, the scaling factors of the plurality of probability models are denoted by α i , a probability of the current bin is p i (t), and entropy obtained with respect to a previous bin is denoted by s i (t−1) of entropy of the current bin, the entropy s i (t) of the current bin is obtained according to an equation S i (t)=bit i *α i +(1−α i )*S i (t−1) by using a parameter bit i obtained according to an equation bit i =(y==1)?−log 2 p i (t):−log 2 (1−p i (t)).
9. An entropy decoding apparatus comprising:
an inverse binarizer configured to map values of predetermined syntax elements to bins of a binary value;
a context modeler configured to receive a predetermined number of bins that are to be binary arithmetic decoded, obtain an autocorrelation value of each bin by using values of the received predetermined number of bins, determine at least one scaling factor used to update a probability of a binary value based on the autocorrelation value, and update a probability used in context-based adaptive binary arithmetic decoding by using the determined at least one scaling factor; and
a regular coder configured to arithmetic decode a current bin by using the updated probability,
wherein the context modeler obtains the updated probability by adding a first value and a second value, the first value being obtained by multiplying a value based on the at least one scaling factor by a probability of a previous bin, and the second value being obtained by multiplying the at least one scaling factor by a value of a current bin,
wherein when a value of the current bin is y (y is 0 or 1), a probability previous to the current bin is p(t−1) (t is an integer), the updated probability is p(t), and the at least one scaling factor is α, the updated probability p(t) is obtained according to an equation P(t)=αy+(1−α)*P(t−1).