This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING 1
A Local Contrast Method for Small
Infrared Target Detection
C. L. Philip Chen, Fellow, IEEE , Hong Li, Member, IEEE, Yantao Wei, Tian Xia, and Yuan Yan Tang, Fellow, IEEE
Abstract—Robust small target detection of low signal-to-noise
ratio (SNR) is very important in infrared search and track appli-
cations for self-defense or attacks. Consequently, an effective small
target detection algorithm inspired by the contrast mechanism
of human vision system and derived kernel model is presented
in this paper. At the first stage, the local contrast map of the
input image is obtained using the proposed local contrast measure
which measures the dissimilarity between the current location and
its neighborhoods. In this way, target signal enhancement and
background clutter suppression are achieved simultaneously. At
the second stage, an adaptive threshold is adopted to segment
the target. The experiments on two sequences have validated the
detection capability of the proposed target detection method.
Experimental evaluation results show that our method is simple
and effective with respect to detection accuracy. In particular, the
proposed method can improve the SNR of the image significantly.
Index Terms—Derived kernel (DK), infrared (IR) image, local
contrast, signal-to-noise ratio (SNR), target detection.
I. INTRODUCTION
I
T IS well known that target detection has found wide
applications in such areas as remote sensing, surveillance,
aerospace, and so on [1]–[4]. As an important technique in
target detection, infrared (IR) imaging has received a lot of
attentions [3], [5]–[7]. Detecting a small IR target of unknown
position and velocity at low signal-to-noise ratio (SNR) is an
important issue in IR search and track system, which is nec-
essary for military applications to warn from incoming small
targets from a distance, such as enemy aircraft and helicopters
[3], [6]–[11]. The target immersed in heavy noise and clutter
Manuscript received September 7, 2012; revised December 12, 2012 and
January 1, 2013; accepted January 2, 2013. This work was supported in
part by the National Natural Science Foundation of China under Grant
61075116, by the Natural Science Foundation of Hubei Province under Grant
2009CDB387, by the Chinese National Basic Research 973 Program under
Grant 2011CB302800, by the Macau Science and Technology Development
Fund under Grant 008/2010/A1, by the Multi-Year Research of University of
Macau under Grants MYRG205(Y1-L4)-FST11-TYY and MYRG187(Y1-L3)-
FST11-TYY, and by the Start-up Research of University of Macau under Grant
SRG010-FST11-TYY.
C. L. P. Chen, T. Xia, and Y. Y. Tang are with the Faculty of Science
and Technology, University of Macau, Macau, China (e-mail: Philip.Chen@
ieee.org; yb17404@umac.mo; yytang@umac.mo).
H. Li is with the School of Mathematics and Statistics, Huazhong
University of Science and Technology, Wuhan 430074, China (e-mail:
hongli@mail.hust.edu.cn).
Y. Wei is with the College of Information Technology, Journalism and
Communications, Central China Normal University, Wuhan 430079, China.
He was with the Institute for Pattern Recognition and Artificial Intelligence,
Huazhong University of Science and Technology, Wuhan 430074, China
(e-mail: yantaowei@mail.ccnu.edu.cn).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TGRS.2013.2242477
background presents as spotlike feature. The gray value of a
target is usually higher than that of its immediate background
in IR image and is not spatially correlated with that of the
local neighborhood. Due to the effects of inherent sensor noise
and natural factors, there exist some high gray regions in the
IR image (e.g., irregular sunlit spot). These effects make the
target detection more difficult. In order to detect a small target
effectively, various algorithms have been developed in the past
few decades [12]–[22].
Conventional small target detection methods such as the me-
dian subtraction filter [12], top-hat filter [13], and max-mean/
max-median filter [14] are widely used to reduce the back-
ground clutters. Genetic algorithm and morphological filters are
also combined to detect the small targets [21]. Image informa-
tion generated by wavelet in different scales supplies the feature
information that could distinguish the target and background.
Based on this idea, many methods of small target detection
in clutter background using wavelet are proposed [23]. Fractal
approaches have been widely used for IR target detection [24].
However, they are sometimes inaccurate and time consuming.
Sun et al. propose target detection methods based on center-
surround difference [19], [20]. In addition, some methods based
on statistical regression have been proposed [17], where the
complex background clutter is predicted and eliminated by a
regression model. Recently, a small target detection method
based on sparse ring representation (SRR) has been proposed
[25]. SRR is an effective graphical structure which can describe
the difference between the background and targets. There are
still many other algorithms for target detection, such as methods
based on manifold learning, empirical mode decomposition,
and neural network [8], [15], [26]. However, small target de-
tection under complex background is still a challenge.
In order to design a small target detection method, it is
inspiring to imitate the computational architecture of biological
vision [27]–[29]. Recently, bioinspired hierarchical algorithms
such as the hierarchical model and X [28], and derived ker-
nel (DK) [30], [31] have attracted much attention. Among
these methods, the DK model studies the theoretical properties
of hierarchical algorithms from a mathematical perspective.
Consequently, the DK model can guide the design of a new
algorithm. Noticed that the small target has a signature of
discontinuity with its neighboring regions and concentrates in a
relatively small region, which can be considered as a homoge-
neous compact region, and the background is consistent with its
neighboring regions [32]. Consequently, we conceive that, after
some target enhancement operation, the local region whose
contrast is larger than the given threshold in some scale may be
a position where the target appears. With these considerations
0196-2892/$31.00 © 2013 IEEE