Image Processing Reference
In-Depth Information
aGrayscaleimage
b Threshold: 50
c Threshold: 70
d Threshold: 100
e Threshold: mean value
f Threshold: median value
Fig. 6.4 Thresholding of a grayscale image
The main problem with this approach is noise sensitivity because adding even a
small amount of noise significantly affects the lower bits. In order to reduce noise
sensitivity, it is possible to use a binary image obtained by some higher bits. For ex-
ample, in Fig. 6.5 proper results can be accomplished by using the image produced
by the 4th significant bit of pixels values (Fig. 6.5e). That binary image is less sen-
sitive to noise, but still contains enough information about image texture. However,
selection of the proper binary level strongly depends on image content. For images
with less homogeneous areas, higher levels could also be used as a proper represen-
tation.
6.4.1.3
Local Binary Pattern
Local Binary Pattern (LBP) is a simple and efficient texture operator obtained by
thresholding the neighbourhood of central pixels and representing the result as a
binary number [16]. It is simple to calculate but also has the important property of
robustness to illumination change.
The LBP of neighbourhood P and radius R is obtained by using the value of
central pixel g c as the threshold for defining values of neighbourhood pixels g p
according to equation (6.5).
1
P
1
p = 0 s ( g p g c ) × 2 p
,
,
x
0
(
,
)=
,
(
)=
LPB
P
R
s
x
(6.5)
0
,
otherwise.
 
Search WWH ::




Custom Search