Multiscale random projection based background suppression of infrared
small target image
Hanlin Qin
a,
⇑
, Jiaojiao Han
a
, Xiang Yan
a
, Jia Li
a,b
, Huixin Zhou
a
, Jingguo Zong
a
, Bingjian Wang
a
,
Qingjie Zeng
a
a
School of Physics and Optoelectronic Engineering, Xidian University, Xi’an 710071, China
b
Science Institute, Air Force Engineering University, Xi’an 710051, China
highlights
We present a background suppression method based on multiscale random projection.
Nonsubsampled pyramid is applied to separates the background and targets.
The proposed method can not only efficiently reduce the spatial redundancy but also preserve target information.
Both background suppression and target enhancement can be achieved by computing Mahalanobis distance in the proposed method.
article info
Article history:
Received 5 July 2015
Available online 13 October 2015
Keywords:
Infrared small target image
Background suppression
Multiscale decomposition
Random projection
abstract
The heavy clutters background in a single infrared image containing small targets is difficult to be effi-
ciently suppressed using the traditional methods. To overcome this difficulty, a novel infrared image
background suppression method based on multiscale random projection is proposed in this paper. On
the one hand, using nonsubsampled pyramid transform that decomposes the original infrared image into
a low-frequency subband and a series of high-frequency subbands, the proposed method separates the
background and targets from the original infrared image. On the other hand, due to a subband image cube
is formed by concatenating together all the high-frequency subbands and a random projection is applied
to the image cube, the proposed method can well preserve target information and reduces the spatial
redundancy of the target and background information. Experimental results on three real infrared image
sequences under different clutters background demonstrate that the proposed method is efficient and
can obtain better performance than other methods for background suppression of infrared small target
images.
Ó 2015 Elsevier B.V. All rights reserved.
1. Introduction
Infrared small target detection is one of the key techniques in
infrared search and track (IRST) system [1]. The detection precision
directly affects the performance of IRST system. If the heavy clut-
ters such as cloud clutters exist in infrared images, the small tar-
gets contained in infrared images will be submerged with the
complex clutters background. What’s more, the small targets lack
the information of concrete shape and texture since the long imag-
ing distance. Thus, small target detection in complex infrared clut-
ters background becomes a difficult and promising problem.
Considering the characteristics of the complex infrared clutters
background discussed above, infrared clutters background sup-
pression has become critical in the process of small target detec-
tion. To do this, researchers have done many correlative research
efforts.
A large number of background suppression methods [2–4] have
been proposed in literatures. These methods can be broadly
grouped into three families [2,5]: spatial filter-based methods,
temporal filter-based methods and transform domain-based meth-
ods. The spatial filter-based methods have a relative long history,
such as the max-meannmax-median filter [6], the morphological
method [7–9], two-dimensional least mean square filter [10], the
bilateral filter [11], bilateral two-dimensional minimum mean
square error [12], background estimation method based on non-
parametric regression [13] and other methods based on biological
vision [14–16]. These methods are based on an assumption that
the background pixels are spatially correlated and the target pixels
http://dx.doi.org/10.1016/j.infrared.2015.09.016
1350-4495/Ó 2015 Elsevier B.V. All rights reserved.
⇑
Corresponding author.
Infrared Physics & Technology 73 (2015) 255–262
Contents lists available at ScienceDirect
Infrared Physics & Technology
journal homepage: www.elsevier.com/locate/infrared