Confidence-driven infrared target detection
Zhiguo Zhang
a
, Liman Liu
b,c
, Wenbing Tao
a,
⇑
, Yuanyan Tang
d
a
School of Automation and National Key Laboratory of Science & Technology on Multi-spectral Information Processing, Huazhong University of Science and Technology,
Wuhan 430074, China
b
School of Biomedical Engineering, South-Central University for Nationalities, Wuhan 430074, China
c
The College of Physical Science and Technology, Central China Normal University, Wuhan 430073, China
d
Department of Computer and Information Science, University of Macau, Macau, China
article info
Article history:
Received 10 January 2014
Available online 11 June 2014
Keywords:
Confidence evaluation
Target detection
Infrared
Localization
False alarm
abstract
The confidence of target detection can be used to evaluate the reliability and risk level of the detected
targets and can effective help to exclude the false alarms, but very little investigation was involved in
the past. In this letter, a confidence-driven infrared target detection method is proposed. We develop
three confidence evaluating methods: (1) the median classification confidence of the cascade classifier;
(2) the context confidence based on the number and the confidence of the merged detection rectangles
around the detected target; and (3) the contrast confidence based on the difference between the detected
target distribution and the around background distribution. The three confidences are combined to form
the final confidence of the detected targets. We then use the confidence to refine the localization of the
targets. The evaluation using real infrared images demonstrates the good performance of the proposed
confidence-driven infrared detection algorithm on both undetected error and false alarm.
Ó 2014 Elsevier B.V. All rights reserved.
1. Introduction
As the development of infrared imaging technology, infrared
target detection has received more and more attention and be
widely applied in automatic target recognition (ATR), especially
in military field due to the robust imaging to the weather situation,
the change of climate and the good performance of all-day working
[1,2]. The infrared target detection is always a challenging task
because of the high variability of target signatures and highly
unpredictable nature of thermal exchange with the environment
[3].
Most of the traditional infrared target detection methods are
based on the difference of the appearance features between the
targets and background. For example, the contrast and the light
feature are usually used to distinguish the targets from the
background [4]. However, due to the serious clutter disturbances
in sea, the significant cloud occlusion in sky or the complex and
varied ground background the feature-based technologies often
fail to extract the true targets or bring about too many false alarms.
It is rather difficult to develop a general approach to detect all
kinds of targets in the different backgrounds [5].
Another noteworthy thing is that the existing infrared target
detection methods are lacking in the confidence evaluation to tar-
get detection. Confidence evaluation can be seen as a variable
which has high correlation with the probability of correct target
detection, and it can be used to measure the reliability of the
detected targets. The extracted targets from images can be sorted
by the order of confidence and more attention should be paid to
the targets with high confidence, which is especially important in
military application. The confidence of target detection can help
to determine the risk level of the detected targets when multiple
targets appear, then decide the defending order of the Early Warn-
ing System (EWS). The confidence can also help to exclude the false
alarms. Therefore, target detection with confidence evaluation is
extremely significant in ATR of military field.
To develop a general infrared target detection which can be
adaptively applied to all kinds of military targets such as airplane,
missile and warships, we resort to the machine learning technolo-
gies which are extensively used to object recognition and classifica-
tion in natural scene [6]. We choose AdaBoost learning algorithm to
build our target detection framework. The AdaBoost learning
algorithm has been successful applied in face detection [7] where
a simple and efficient classifier is built to select a small number of
critical visual features from a very large set of potential features
[8] to achieve high detection rates. Simultaneously, two strategies,
the integral image representation and cascade combination of
http://dx.doi.org/10.1016/j.infrared.2014.05.015
1350-4495/Ó 2014 Elsevier B.V. All rights reserved.
⇑
Corresponding author. Tel.: +86 27 87541924.
E-mail address: wenbingtao@hust.edu.cn (W. Tao).
Infrared Physics & Technology 66 (2014) 78–83
Contents lists available at ScienceDirect
Infrared Physics & Technology
journal homepage: www.elsevier.com/locate/infrared