New technique for the visualization of high
dynamic range infrared images
Francesco Branchitta
Marco Diani,
MEMBER SPIE
Giovanni Corsini
University of Pisa
Dipartimento di Ingegneria dell’Informazione
Via G. Caruso 12
Pisa, 56124
Italy
E-mail: m.diani@iet.unipi.it
Marco Romagnoli
Selex-Galileo
Via Albert Einstein 35
Campi Bisenzio Firenze, 50013
Italy
Abstract. We propose a new dynamic range compression technique for
infrared 共IR兲 imaging systems that enhances details visibility and allows
the control and adjustment of the image appearance by setting a number
of tunable parameters. This technique adopts a bilateral filter to extract a
details component and a coarse component. The two components are
processed independently and then recombined to obtain the output-
enhanced image that fits the display dynamic range. The contribution
made is threefold. We propose a new technique for the visualization of
high dynamic range 共HDR兲 images that is specifically tailored to IR im-
ages. We show the effectiveness of the method by analyzing experimen-
tal IR images that represent typical area surveillance and object recog-
nition applications. Last, we quantitatively assess the performance of the
proposed technique, comparing the quality of the enhanced image with
that obtained through two well-established visualization methods.
© 2009
Society of Photo-Optical Instrumentation Engineers. 关DOI: 10.1117/1.3216575兴
Subject terms: infrared images; dynamic range compression; image
enhancement
.
Paper 090052R received Jan. 23, 2009; revised manuscript received Jun. 26,
2009; accepted for publication Jul. 13, 2009; published online Sep. 4, 2009.
1 Introduction
Modern thermal cameras have high dynamic range 共HRD;
up to 14 bits兲 and can potentially accommodate large signal
variations without losing the perceptibility of low-contrast
details. Unfortunately, traditional display devices have only
256 gray levels and cannot represent all the information
available in the original digital data. Compression of the
input signal to the 256 digital levels available on the
display—for example, by simple linear mapping—may
lead to a dramatic reduction of the contrast associated to
important details that often represent the relevant informa-
tion sought in the scene. Therefore, more complex tech-
niques are required in order to adapt the recorded signal to
the display, maintaining, and possibly improving, the image
visual quality. In this work, we propose a new visualization
technique, specifically tailored to thermal images, whose
goal is to give to the observer a natural perception of the
monitored scene and, at the same time, enhance the local
contrast to improve details visibility. The technique has
been designed so as to take into account some of the main
requirements of typical thermal image applications. Spe-
cifically, an effective visualization technique should lead to
• Compression of large gradients
• Enhancement of the visibility of low-contrast details
• A pleasant and natural 共i.e., easy to interpret for a
human operator兲 representation of the scene, avoiding
the introduction of artifacts
The problem considered in this work includes aspects that
are typical of two different research fields widely investi-
gated in the literature: dynamic range compression 共DRC兲
and tone mapping 共TM兲. DRC techniques aim at forcing the
data to the dynamic range of the display while enhancing
local contrast.
1–3
Instead, the goal of TM techniques is to
visualize an image that reproduces as faithfully as possible
what a viewer would perceive when observing the scene.
4,5
Existing methods have been derived for images acquired
in the visible spectral domain, and the problem remains
open in the area of infrared 共IR兲 imaging devices,
6
where
finding the trade-offs between giving a correct perception
of the global scene and improving the perceptibility of the
details is not a trivial task.
Among DRC techniques, unsharp masking 共UM兲 meth-
ods play an important role. UM methods use linear filters to
separate the low spatial frequency 共LSF兲 and the high spa-
tial frequency 共HSF兲 components of the image. Then, a
fraction of the HSF image is added to the original image to
obtain the sharpening effect. Last, the dynamic range of the
output image is linearly compressed. As a result, dynamic
compression is accomplished, and at the same time, detail
visibility is enhanced through the increased emphasis given
to the highest spatial frequencies. UM techniques suffer
from overenhancement of the HSF components, which pro-
duces artifacts 共called halos兲 in those regions where large
intensity variations are present.
Since the human eye itself performs a compression of
the range of luminance, tone mapping algorithms are also
related to the problem of DRC. The simplest tone mapping
techniques adopt nonlinear mapping functions such as loga-
rithmic, gamma, sigmoid, and knee curves.
1,2,7
To reduce the halo artifacts that affect UM techniques,
many nonlinear filters have been proposed. In 1999, Tum-
blin and Turk
8
proposed the nonlinear anisotropic diffusion
共AD兲 algorithm. This technique is based on a principle
similar to that which regulates heat exchange between two
0091-3286/2009/$25.00 © 2009 SPIE
Optical Engineering 48共9兲, 096401 共September 2009兲
Optical Engineering September 2009/Vol. 48共9兲096401-1
Downloaded from SPIE Digital Library on 25 Dec 2010 to 117.32.153.160. Terms of Use: http://spiedl.org/terms