Information Technology Reference
In-Depth Information
gradient features with a LBP-based feature. The CS-LBP feature has a rel-
atively short feature histogram, is tolerance to illumination changes and is
computational simple. The performance of the CS-LBP descriptor was com-
pared to that of the SIFT descriptor in the contexts of matching and object
category classification [8]. For many tests, the proposed CS-LBP method out-
performs the SIFT descriptor, while in the other cases, the performance is
comparable to SIFT [8].
Recently, LBP has been recently modified in order to be used in the context
of dynamic texture description and recognition with also an application to
facial expression analysis by Zhao et al. [22]. They introduced the extension
of the LBP operator into the temporal domain, named Volume Local Binary
Pattern (VLBP), and a different and simplified version which consider the co-
occurrences on Three Orthogonal Planes named LBP-TOP. The basic VLBP
labels a volume thresholding a neighborhood region not only in the current
frame but also in previous and following frames and encoding the results as
a binary number. The drawback of this method is that a large number of
neighborhood
P
2
3
P
+2
) while a small
P
means losing information. LBP-TOP makes the approach computationally
simpler and easier extracting the LBP code from three orthogonal planes (XY,
XT and YT) denoted as XY-LBP, XT-LBP and YT-LBP. In such a scheme,
LBP encodes appearance and motion in three directions, incorporating spatial
information in XY-LBP and spatial temporal co-occurrence statistics in XT-
LBP and YT-LBP.
LBP-TOP computes the LBP from Three Orthogonal Planes, denoted as
XY-LBP, XT-LBP and YT-LBP. The operator is expressed as
produces a very long feature vector (
LBP − TOP
P
XY
,P
XT
,P
YT
,R
X
,R
Y
,R
T
(5)
where the notation
P
XY
,P
XT
,P
YT
,R
X
,R
Y
,R
T
denotes a neighborhood of
P
points equally sampled on a circle of radius
R
on
XY
,
XT
and
YT
planes
respectively. The statistics on the three different planes are computed and
then concatenated into a single histogram as
x,y,t
I
(
f
j
(
x, y, t
)=
i
)
,
H
i
=
i
=0
,
1
, ..., n
j
;
j
=1
,
2
,
3
(6)
where
n
j
is the number of different labels produced by the LBP operator
in the jth plane,
fj
is the central pixel at coordinates
(
x, y, t
)
in the
j
th
plane and
I
(
A
)
is 1 if
A
is true and 0 otherwise.The resulting feature vector
P
length. Fig.6 illustrates the construction of the LBP-TOP de-
scriptor.(( In such a scheme, LBP encodes appearance and motion in three
directions, incorporating spatial information in XY-LBP and spatial tempo-
ral co-occurrence statistics in XT-LBP and YT-LBP)).
In our implementation, LBP-TOP is applied on each cuboid and XY, XT
and YT planes are the central slices of it, as shown in Figure 6 and in
Figure 7, in which the cuboid is extracted from the running sequence. The
is of
3
·
2
Search WWH ::
Custom Search