with the background vegetation. The change detection of
INV_
NDVI
is done by subtracting inverse
NDVI
before the landslide
from inverse
NDVI
after the landslide, and the results for the
ten research regions are demonstrated in Figure 6.
Saliency Probability Calculation
On top of the change detection of inverse
NDVI
, the areas where
landslide occurs (shown in Figure 4) are highlighted. We
employed the FASA method (Yildirim and Süsstrunk, 2014)
to extract the outstanding regions, which are called salient re-
gions. Salient regions are regions that attract our interest most
when we stare at an image at first sight, which are determined
by color, texture, and spatial characteristics of an image.
FASA, a fast, accurate and size-aware salient object detection
method, is an outstanding method in saliency object detection
for its highly efficiency. The general description is as follows:
Color Space Transformation
CIEL*a*b* is a commonly used color domain to describe all the
colors that human eyes can visualize. It is proposed by the In-
ternational Commission on Illumination, Commission Interna-
tionale d’Eclairage (
CIE
) (Jedi, 2008) and has been widely used
in image segmentation and object detection (Woelker, 1996).
Calculate Spatial Center and Variances of a Color
For each pixel (
x
i
,
y
i
), two vectors
P
i
and
C
i
and are calculated
according to Equation 3:
P
x
y
C
L P
a P
b P
i
i
i
i
i
i
i
=
=
*
*
*
( )
( )
( )
(3)
where
L
*(
P
i
),
a
*(
P
i
),
b
*(
P
i
) are color elements of pixel (
x
i
,
y
i
) in
CIEL*a*b* color space. Based on
P
i
and
C
i
, the spatial center
{m
x
(P
i
), m
y
(P
i
)} and variances {V
x
(P
i
), V
y
(P
i
)} of color
C
i
in verti-
cal direction are calculated as Equations 4 and 5:
m P
w C C y
w C C
y i
c
i
j
j
j
N
c
i
j
j
N
( )
( , )
( , )
=
⋅
=
=
∑
∑
1
1
(4)
V P
w C C y m P
w C C
y i
c
i
j
j
y i
j
N
c
i
j
j
N
( )
( , ) (
( ))
( , )
=
⋅
−
=
=
∑
∑
2
1
1
(5)
(a)
(b)
(c)
(d)
(e)
(f)
Figure 6. Change detection of inverse NDVI of six research regions.
Continued on next page.
356
May 2017
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
