572 CHINESE OPTICS LETTERS / Vol. 7, No. 7 / July 10, 2009
Bridge recognition of median-resolution SAR images using
pun histogram entropy
Wenyu Wu (吴吴吴文文文宇宇宇)
∗
, Dong Yin (尹尹尹 东东东), Rong Zhang (张张张 荣荣荣),
Yan Liu (刘刘刘 岩岩岩), and Jia Pan (潘潘潘 嘉嘉嘉)
Institute of Image Processing and Remote Sensing, University of Science and Technology of China,
Hefei 230027, China
∗
E-mail: wenyuwu@mail.ustc.edu.cn
Received August 13, 2008
A novel algorithm for bridge recognition of median synthetic aperture radar (SAR) images using histogram
entropy presented by Pun is proposed. Firstly, Lee filter and histogram proportion are used to denoise
the original image and to make the target evident. Then, water regions are gained through histogram
segmentation and the contours of water regions are extracted. After these, the potential bridge targets are
obtained based on the space relativity between bridges and water regions using improved contour search.
At last, bridges are recognized by extracting the feature of Pun histogram entropy (PHE) of these potential
bridge targets. Experimental results show the good qualities of the algorithm, such as fast speed, high rate
of recognition, and low rate of false target.
OCIS codes: 100.0100, 100.3008, 280.6730.
doi: 10.3788/COL20090707.0572.
Synthetic aperture radar (SAR) has been widely ap-
plied to gain large-area and high-resolution images in
aerospace, ground reconnaissance, remote sensing or re-
source census, etc. Target recognition in SAR image has
been one of the hotspots in the development of remote
sensing. Bridge recognition is one important kind of SAR
image target recognition, as the recognition is very useful
for image registration, precision-guidance, map drawing,
target detection, and so on.
Bridge targets have some distinct features in median-
resolution SAR images, such as the high gray value, the
limited length and width, the usually constant width,
the changeless direction, the approximately straight line
edges, and water regions which have constantly low gray
value are usually besides them, etc. By now, the ways
for SAR image bridge recognition are classified into three
aspects. One is based on amalgamation between SAR
images and optical images
[1−8]
, which synthesizes the in-
formation of each kind of images to recognize the tar-
gets. Another is based on morphological method
[2]
,
which makes water regions connected and recognizes the
targets in these connected regions. The third one is based
on edge information extraction
[3−9]
, which detects the
contours in the images and gains the result using the
features of target contours, such as parallel edges. The
first aspect needs different kinds of images of one region,
which is hard to realize, although it has high precision.
The second aspect cannot be automatic, for the morpho-
logical disposals are not the same in different images.
The third aspect can cause high rate of false target, for
the edges in SAR images are rugged and the noise has
severe interference. Even more, sometimes the edges are
not parallel in median-resolution SAR images. In this let-
ter, using space relativity between bridges and water re-
gions and the feature of Pun histogram entropy (PHE)
[6]
,
we propose a novel algorithm for bridge recognition of
median-resolution SAR images, in which we use the fea-
ture of PHE instead of sharp features to make sure that
the target can be recognized even if its shape is distorted
by noise.
For the noise in SAR images is usually multiplicative,
the difference operator which is used in optical images
should not be used here. Instead, there are many other
filters to denoise it, such as Lee filter
[4]
, Frost filter, Kuan
filter
[5]
, and independent component analysis (ICA)
[10]
,
etc. Lee filter
[4]
, which is based on the model of complete-
grown multiplicative fleck noise, is used in this letter ac-
cording to the following consideration. Although the Lee
filter results in some fuzzy images, the edge contours of
water are preserved and the features of land targets are
weakened.
Suppose that transcendent mean and variance can be
gained by partial mean and variance, then
R = I + (C
u
− I) × W (t) , (1)
W (t) = 1 − C
2
u
/C
2
I
, (2)
where C
I
>C
u
, R is the partial gray value after smooth-
ing process. I, σ
I
, and C
u
are the mean of partial gray
value, the standard variance of image data, and the gray
value of the center pixel of the partial area in smoothing
model, respectively. C
I
is the partial variance parameter
of I, C
I
=σ
I
/I. Using Eq. (1), the image smoothing is ac-
tually performed so that the pre-processed image would
be used in the next step.
Sharp features are usually used in bridge recognition
of SAR images. Admittedly, sharp features are visual
and easily extracted. However, the noise interference and
the imaging mechanism give rise to the distortion of tar-
get sharps, causing that sharp features are unable to de-
scribe the universal characteristics of different bridges in
different images. The PHE feature, used in this letter
shows the distribution of gray levels and the abundance
of information, and almost not changed by sharp distor-
tion. Therefore, it can be widely used in different images.
1671-7694/2009/070572-04
c
° 2009 Chinese Optics Letters