A sparse representation-based method for infrared dim target detection
under sea–sky background
Xin Wang
a,b,
⇑
, Siqiu Shen
a
, Chen Ning
c
, Mengxi Xu
d
, Xijun Yan
a
a
College of Computer and Information, Hohai University, Nanjing, Jiangsu 211100, China
b
Jiangsu Key Laboratory of Image and Video Understanding for Social Safety, Nanjing University of Science and Technology, Nanjing, Jiangsu 210094, China
c
School of Physics and Technology, Nanjing Normal University, Nanjing, Jiangsu 210023, China
d
School of Computer Engineering, Nanjing Institute of Technology, Nanjing, Jiangsu 211167, China
highlights
A new sparse representation-based method is presented for infrared dim target detection.
An overcomplete background dictionary is learned for cluttered sea–sky background.
Target candidates are predicted via sparse reconstruction errors.
Two constraints are combined together for dim IR target identification.
article info
Article history:
Received 24 February 2015
Available online 29 May 2015
Keywords:
Infrared image
Target detection
Sparse representation
Dictionary learning
abstract
Automatic detection for infrared (IR) dim targets under complex sea–sky background is a challenging
task. To explore an effective solution to the problem, this paper develops a sparse
representation-based method by learning a sea–sky background dictionary. This framework is mainly
composed of three modules: background dictionary learning, preliminary target localization, and accu-
rate target identification. In the first module, a sea–sky background dictionary is learned from a large
number of training samples, which has a good ability to model the cluttered sea–sky background. In
the second module, given a test image, it is first divided into a set of patches; then, for each image patch,
its sparse representation coefficients are computed over the learned dictionary. By analyzing the sparse
reconstruction errors for the image patches, the target candidate areas can be predicted. In the third mod-
ule, an infrared dim target recognition scheme is applied to those areas to recognize the true dim IR tar-
gets. Based on a set of comprehensive experiments, our algorithm has demonstrated better performa nce
than several other infrared dim target detection methods.
Ó 2015 Elsevier B.V. All rights reserved.
1. Introduction
Infrared (IR) dim target detection under complicated sea–sky
background is a key technology for a wide range of applications
such as security surveillance, navigation, defense, etc. With the
development of infrared imaging technology, infrared sensors have
provided high resolution images, which facilitate the detection of
targets. However, it is still a challenging problem to reliably detect
dim IR targets with complex sea–sky background not only due to
lack of valid appearance information, such as shape, texture, color,
etc, of the objects, but also because of complex background clutter
[1–3].
Infrared dim target detection tasks often start from predicting
background or filtering the target signals from background pixels.
For example, Gu et al. [4] proposed a kernel-based nonparametric
regression method for background prediction and small target
detection in infrared images. Dong et al. [5] developed a homoge-
neous background prediction model for detection of infrared point
target. Zeng et al. [6] used a Top-Hat morphological filter for infra-
red image target detection. Bae et al. [7] introduced an edge direc-
tional two-dimensional least squares filter to detect a small target
in infrared images. In addition, other researchers have applied frac-
tal theory for infrared small objects detection [8,9] based on the
fact that fractal dimension between man-made objects and the
natural background is different. Recently, visual attention
http://dx.doi.org/10.1016/j.infrared.2015.05.014
1350-4495/Ó 2015 Elsevier B.V. All rights reserved.
⇑
Corresponding author at: College of Computer and Information, Hohai Univer-
sity, Nanjing, Jiangsu 211100, China.
E-mail address: wang_xin@hhu.edu.cn (X. Wang).
Infrared Physics & Technology 71 (2015) 347–355
Contents lists available at ScienceDirect
Infrared Physics & Technology
journal homepage: www.elsevier.com/locate/infrared