Fast motion estimation methods using multiple reference frames
Motion estimation methods using multiple reference frames are provided. In the methods, motion estimation is performed on a current block using a single reference frame selected from the multiple reference frames or using the multiple reference frames in order of probability that each reference frame is an optimal reference frame for the current block. Accordingly, faster motion estimation is performed.
1. A motion estimation method using a plurality of reference frames, the method comprising: analyzing a plurality of reference indexes of a current block, the plurality of reference indexes indicating reference frames of neighboring blocks that have already been coded and a plurality of reference indexes of a block corresponding to the current block and its neighboring blocks in a previous frame; performing a first linear combination of a number of first blocks that have a same reference index among the neighboring blocks of the current block, and a number of second blocks that have the same reference index among the block corresponding to the current block and its neighboring blocks in the previous frame; estimating a norm of a motion vector of the current block based on a norm of an average motion vector of the neighboring blocks of the current block; calculating a posterior probability of the estimated norm of the motion vector of the current block with respect to each of the plurality of reference frames using the estimated norm of the motion vector of the current block as a feature vector; performing a second linear combination of a value of the first linear combination and the posterior probability of the estimated norm of the motion vector of the current block with respect to each reference frame and performing motion estimation on the current block sequentially using the plurality of reference frames according to values of the second linear combination; and calculating a probability distribution of the reference frame index using an equation:
P
(
i
X
;
X
)
=
P
X
(
i
X
)
=
e
-
a
(
i
X
+
1
)
∑
j
=
0
r
-
1
e
-
a
(
j
+
1
)
∑
k
=
0
r
-
1
P
X
(
i
x
)
=
∑
k
=
0
r
-
1
P
X
(
i
X
=
k
)
P
^
k
(
N
)
+
∑
k
=
0
r
-
1
P
X
(
i
X
=
k
)
P
^
k
C
(
N
)
=
Γ
N
Γ
N
C
,
for
i
X
∈
R
,
R
=
{
0
,
…
,
r
-
1
}
,
a
>
0
,
.
2. The motion estimation method of claim 1 , wherein performing the first linear combination comprises multiplying the number of first blocks and the number of second blocks by different weights, respectively; and
wherein a weight for the number of first blocks is greater than a weight for the number of second blocks.
3. The motion estimation method of claim 1 , wherein performing the second linear combination and performing the motion estimation comprise multiplying the value of the first linear combination and the posterior probability by different weights, respectively; and
wherein the weight for the second linear combination is greater than the weight for the posterior probability.
4. The motion estimation method of claim 1 , wherein estimating the norm of the motion vector of the current block comprises:
generating a plurality of training samples representing relations between a norm of a measured motion vector of the current block and a norm of an average motion vector of the neighboring blocks; and
estimating a norm of an average motion vector of the neighboring blocks that corresponds to a training sample having a minimum error among the plurality of training samples, as the norm of the motion vector of the current block.
5. The motion estimation method of claim 1 , wherein calculating the posterior probability comprises:
calculating a prior probability of the estimated norm of the motion vector of the current block with respect to each reference frame;
calculating a probability density function of the estimated norm of the motion vector of the current block with respect to the reference frame; and
calculating the posterior probability of the estimated norm of the motion vector of the current block with respect to the reference frame based on the prior probability and the probability density function.
6. A motion estimation method using a plurality of reference frames, the method comprising: analyzing a plurality of reference indexes of a current block, the plurality of reference indexes indicating reference frames of neighboring blocks that have already been coded and a plurality of reference indexes of a block corresponding to the current block and its neighboring blocks in a previous frame; performing a first linear combination of a number of first blocks that have a same reference index among the neighboring blocks of the current block, and a number of second blocks that have the same reference index among the block corresponding to the current block and its neighboring blocks in the previous frame; estimating a norm of a motion vector of the current block based on a norm of an average motion vector of the neighboring blocks of the current block; calculating a posterior probability of the estimated norm of the motion vector of the current block with respect to each of the plurality of reference frames using the estimated norm of the motion vector of the current block as a feature vector; performing a second linear combination of a value of the first linear combination and the posterior probability of the estimated norm of the motion vector of the current block with respect to each reference frame and performing motion estimation on the current block sequentially using the plurality of reference frames according to values of the second linear combination; and calculating the posterior probability of the estimated norm of the current block's motion vector using an equation:
MP
[
ref
_
idx
]
=
P
(
NMV
|
ref_idx
)
·
P
(
ref_idx
)
Max
{
P
(
NMV
|
k
)
·
P
(
k
)
}
,
for
ref_idx
,
k
∈
{
0
,
1
,
…
,
r
-
1
}
.