AUTOMATIC BINARIZATION METHOD IN ISAR IMAGE
Xiaohui Zhao, Yicheng Jiang, and Yun Zhang
Research Institute of Electronic Engineering Technology,
Harbin Institute of Technology Harbin, Harbin, 150001, China
ABSTRACT
One of the most important image processing technique is the
image segmentation technique, which is normally achieved
by finding the proper threshold value for segmenting the
images. This paper proposes an automatic binarization
method for 2D radar images. The essential idea is to
evaluate the derivative of the cumulative histogram of the
given image and then find the inflection point, which
indicates the proper binarization value for the given image.
The proposed method is compared with a representative
binarization method and the experimental results show the
effectiveness of the proposed method.
Index Terms—Image segmentation, threshold selection,
foreground segmentation, synthetic aperture image, inverse
synthetic aperture image.
1. INTRODUCTION
Automatic binarization is one of the most essential
processing for different kinds of images. The aim is to find a
threshold to separate the given image into two parts, one part
with pixel values higher the threshold and another part with
pixel values lower than the threshold. With proper selection
of the threshold, the target-of-interest can be segmented
from the given image [1]. Many methods [2 - 7] have been
proposed in the fields of computer vision, such as defect
detection [2], document binarization [3], etc..
The methods of threshold selection for binarization
mainly fall into two categories: global binarization [2, 5, 7,
8] and local binarization [1, 3, 4, 6, 9, 10]. Global
binarization methods find a single threshold for the whole
image while local binarization methods find different
thresholds for different regions of the image. The local
binarization methods are actually more effective for non-
uniform illumination images. In fact, the image contains only
the scatter points and noise where non-uniform illumination
is not the main issue. Therefore, global binarization is more
suitable for the ISAR image. Moreover, global binarization
methods are easier to implement with less computational
burden. However, most existing methods are specialized in
other applications, and cannot achieve the optimal threshold
in the ISAR image. This is because the ISAR image is
mainly generated from the echo of the target scatters, the
side-lope of strong scatters and the speckle noise. Moreover,
comparing with other kinds of images, the ISAR image is
sparsely distributed. The aim of the binarization in the ISAR
image is actually finding the threshold separates the target
scatters from the side-lope of strong scatters and the speckle
noise. A well segmented ISAR image is essential for
subsequent processes [11]. In this paper, we propose a
binarization method specialized for the ISAR image.
Experiments based on real ISAR data and the comparison
with the representative automatic binarization method (the
Otsu’s method) show the effectiveness of the proposed
method.
The rest of this paper is organized as follows. The proposed
method is introduced in Section II, followed by experimental
results in Section III. Finally, conclusions are drawn in
Section IV.
2. PROPOSED BINARIZATION METHOD
The essence of binarization is finding a proper threshold to
separate the foreground (target-of-interest) from the
background. Since ISAR images are generated from the
scatters echoes, the generated images are mainly constitute
only of the target-of-interest scatters, their side lopes and the
background noise. Therefore, the aim of binarization in
ISAR image is finding the optimal threshold that separate
the target-of-interest scatters from the rest subjects. Since
the variation of scatters is smooth from the side lope to the
main lope, it is not easy for existing automatic binarization
methods finding the proper threshold.
The histogram and the cumulative histogram are
employed for analyzing the given image. A histogram is an
array
1
{m }
k
ii
that counts the grayscale of image pixels fall
into each bins. A cumulative histogram counts cumulative
number of image pixels in all of the bins up to a specified
bin. Let
n
be the number of image pixels and
k
is the
number of bins. A histogram
1
{m }
k
ii
meets the following
condition:
5415978-1-4799-7929-5/15/$31.00 ©2015 IEEE IGARSS 2015