Thermal Image Enhancement Algorithm Using Local And Global Logarithmic
Transform Histogram Matching With Spatial Equalization
Viacheslav Voronin
Lab. "Mathematical methods of image processing and
intelligent computer vision systems"
Don State Technical University
Rostov-on-Don, Russian Federation
voroninslava@gmail.com
Svetlana Tokareva
Lab. "Mathematical methods of image processing and
intelligent computer vision systems"
Don State Technical University
Rostov-on-Don, Russian Federation
tokareva@bk.ru
Abstract—This paper presents a new thermal image
enhancement algorithm based on combined local and global
image processing in the frequency domain. The presented
approach uses the fact that the relationship between stimulus
and perception is logarithmic. The basic idea is to apply
logarithmic transform histogram matching with spatial
equalization approach on different image blocks. The resulting
image is a weighted mean of all processing blocks. The weights
for every local and global enhanced image driven through
optimization of measure of enhancement (EME). Some
presented experimental results illustrate the performance of
the proposed algorithm on real thermal images in comparison
with the traditional methods.
Keywords-thermal image; infrared image; enhancement;
equalization; logarithmic transform
I. INTRODUCTION
Image enhancement is the important tool in image
processing with a purpose to improve the visual appearance
of the image or to provide “better” transform for future
analysis (segmentation, contour detection and recognition)
[1, 2]. In many cases, enhancement procedures are used as a
preprocessing step for image visualisation.
Infrared and thermal images have been used widely in the
different security applications [3]. Such images show the
temperature difference between different objects and scene
background. One of the drawbacks of such images is low
contrast and noisy images which should be enhanced. This is
a serious problem of infrared and thermal images because of
the character of IR detectors [4].
Enhancement approach can be classified into two
categories: spatial and frequency domain methods [5].
The first group use spatial domain image processing
which directly manipulates the pixels. Many spatial image
978-1-5386-6568-8/18/$31.00 ©2018 IEEE
Authorized licensed use limited to: Cornell University Library. Downloaded on September 02,2020 at 11:28:31 UTC from IEEE Xplore. Restrictions apply.
Evgenii Semenishchev
Lab. "Mathematical methods of image processing and
intelligent computer vision systems"
Don State Technical University
Rostov-on-Don, Russian Federation
sea.sea@mail.ru
Sos Agaian
Dept. of Computer Science
CUNY/The College of Staten Island
Staten Island, New York, United States
sos.agaian@csi.cuny.edu
enhancement methods are based on histogram analysis and
modification; other methods are based on the local contrast
transformation [1].
The most popular image enhancement method is
histogram equalization. It is a global processing approach, so
the entire tone of the image has been changed like more
bright or dark image. In many cases, these methods extend
the dynamic range of an image in local regions, leading to
artefacts and overall tonal change of the image.
The second group use transformation in the frequency
domain through modification magnitudes and altering the
frequency content of the image. These enhancement
techniques use frequency transform such as DCT, Fourier,
etc. Sometimes the image properties such as low and high-
frequency coefficient’s histograms may be so tightly packed
that distinguishing them from one another may be impossible
[1, 6]. Logarithmic transform allows improving the
difference between levels of images [1].
Adaptive histogram equalization (AHE) is an image
processing technique used to improve contrast in images [7,
8]. An adaptive version of this algorithm called contrast
limited adaptive histogram equalization (CLAHE) [8,9]. This
approach is wildly used for improving the local contrast for
thermal and infrared images.
Other effective thermal and IF image enhancement
method base on wavelet transformation and Retinex has been
proposed in [10]. A method based on discrete stationary
wavelet transform (DSWT) has been proposed in [11]. The
enhancement quality in these methods is not natural for
many real type scenes and may be improving for a thermal
image with the following properties: irregular lighting and
brightness gradient.
Each of these methods has strong and weak points.
Hence, the combination of the above methods is used to
5 SSIAI 2018
enhance the image through transform histogram mapping We perform image transformation, which needs to be
technique [12]. enhanced, then manipulated the transform coefficient, and
In this paper, novel enhancement technique for thermal then perform the inverse orthogonal transform.
and infrared images is investigated. The proposed In the proposed approach the Discrete Fourier Transform
contribution is based on logarithmic transform coefficient (DFT) is used. The histogram of this data is usually compact
histogram mapping technique with adaptive histogram and uninformative. Logarithmic transform in our scheme
equalization on different image blocks. helps to maps a narrow range of low grey level values into a
wider range of the output level. This type of transformation
II. PROPOSED ALGORITHM is used to expand the values of dark pixels in an image while
The block diagram of the proposed enhancement compressing the higher-level values [13].
algorithm for the thermal image is depicted in Fig. 1: The Log transformation includes two steps:
- the creation of a matrix to preserve the phase of the
Input Image
transformed image, which is given by the equation:

Image Splitting
(i, j )  angle( X (i, j )) .

The angle function returns the angle of the coefficient

Sub Images Sub Images Sub Images and will be used to restore the phase of the transform
(8 by 8) (16 by 16) (32 by 32)
coefficients.
The transformation of the coefficients according to the
Enhancement Processing
following equation:

Xˆ (i, j )   ln( X (i, j )   ) ,

Enhanced Enhanced Enhanced Enhanced
Image 1 Image 2 Image 3 Image 3

where η, γ, and λ are enhancement parameters, usually set to

EME EME EME EME 1.
calculation calculation calculation calculation To return the coefficients to the standard transform
domain the signal is exponentiated, and the phase is restored
Weighted Average
as shown by the following equation:

~ ˆ
X (i, j )  e X (i , j )  e j(i , j ) .
Enhanced Image

Figure 1. Block diagrams of the proposed algorithm. This allows to keep the overall image characteristics and
return them to the original image.
The procedure for the proposed algorithm is expressed as The drawback of the original adaptive histogram
following steps: equalization method is choosing the optimal size of regions.
Input - Original Image. In the most cases, it depends on the type of the input image.
Step 1 – Image splitting. Therefore, we propose to use CLANE on every disjoint
We split image in moving windows on disjoint blocks block with different sizes (8 by 8, 16 by 16, 32 by 32 and,
with different sizes (8 by 8, 16 by 16, 32 by 32 and, i.e.) i.e.) in moving windows.
(Fig. 2). Histogram mapping is a more generalised version of
8 16 32
8
16
histogram equalization which allows us to specify the shape
32 of the histogram that we wish the processed image to have
[14]. The method used to generate a processed image that
has a specified histogram is called histogram matching or
histogram specification.
Histogram Orthogonal
Figure 2. Image splitting. Equalisation Transform
Log Transform

Step 2 – Enhancement Processing. Input Image

For every sub-image, we use the frequency domain Orthogonal

Transform
Orthogonal
Transform
Log Transform Match Histogram

enhancement method based on the logarithmic transform

histogram matching with spatial equalization. The block Inverse Orthogonal Inverse Orthogonal Inverse Log
Enhanced Image
diagram of the enhancement processing shows in Figure 3. Transform Transform Transform

Figure 3. Block diagram of the enhancement processing.

Authorized licensed use limited to: Cornell University Library. Downloaded on September 02,2020 at 11:28:31 UTC from IEEE Xplore. Restrictions apply.
 This method includes three steps: III. EXPERIMENT RESULTS
1) Histogram equalize the original image. In this section, we compare our results with well-known
2) Histogram equalize the output image. algorithms histogram equalization and CLAHE on the LTIR
Histogram equalization G (z ) calculates by the following dataset v1.0 [18]. This dataset contains 20 thermal infrared
z sequences (garden, horse, trees, street, car et al.).
equation G ( z )  p (t )dt .
 z Figures 4-6 demonstrate the thermal image enhancement
0 results obtained by various algorithms respectively (a –
3) The inverse of the second transform to the original original image; b - the enhanced image by the histogram
equalized image. equalization; c - the enhanced image by the CLAHE; d - the
For histogram equalization G (z ) should be equal to enhanced image by the proposed method). The results
T (r ) , i.e. z  G T (r ) .
1 achieved by current proposed scheme have visually more
contrast.
Step 3 – EME calculation.
There are several methods introduced as a measure of
image enhancement [15-18]. To measure the quality (or
contrast) of images and select the optimal processing
parameters, we use the following quantitative measure of
image enhancement proposed in [14]. The EME is image
enhancement measure introduced by Agaian in [16]:

1 k1 k2 X
EMEk1 , k2  max(   20  log X max;
k1  k2 l 1 k 1 
k ,l
)
a) b)
min; k ,l
 
where X max; k ,l and X min; k ,l respectively are the
minimum and maximum of the image x ( n, m) inside the
block  k ,l .
We calculate EME for every enhanced image:
~
- for enhanced image 1 (all image) is EME X 1 ;
c) d)
~ Figure 4. Thermal Image Enhancement.
X2
- for enhanced image 2 (blocks 8 by 8) is EME ;
- for enhanced image 3 (blocks 16 by 16) is
~
EME X 3 ;
- for enhanced image 4 (blocks 32 by 32) is
~
EME X 4 .
These values allow to calculate weights, as follow:
~
~
X1 EME X 1 a) b)
W  ~ ~ ~ ~ ;
EME  EME  EME  EME
X1 X2 X3 X4

~
~
X2 EME X 2
W  ~ ~ ~ ~ ;
EME X 1  EME X 2  EME X 3  EME X 4
~
~ X3
EME
W X3  ~ ~ ~ ~ ;
EME X 1  EME X 2  EME X 3  EME X 4
~ c) d)
~ X4 Figure 5. Thermal Image Enhancement.
X4 EME
W  ~ ~ ~ ~ .
EME X 1  EME X 2  EME X 3  EME X 4 Proposed approach shows more details in the obtained
Step 4 – Weighted Average. enhanced thermal images. The analysis shows, what the
The resulted enhanced images define as: proposed method gives better visual quality than histogram
~ ~ ~ ~ ~ ~ ~ ~ ~ equalization technique and CLANE.
X  X 1W X 1  X 2 W X 2  X 3 W X 3  X 4 W X 4 . The experimental results (Table 1) show that the original
Output – Enhanced Image. images have the lowest EME. After applying proposed local
and global processing, the EME has risen. It is noticeable

Authorized licensed use limited to: Cornell University Library. Downloaded on September 02,2020 at 11:28:31 UTC from IEEE Xplore. Restrictions apply.
that the quality of the obtained results by the proposed
algorithm has several times better in regard to the EME
measurement. REFERENCES
[1] A. Grigoryan and S. Agaian, “Image enhancement,” in Advances in
Imaging and Electron Physics. New York: Academic, pp. 165–243,
2004.
[2] J. Tang, E. Peli, and S. Acton, “Image enhancement using a contrast
measure in the compressed domain,” IEEE Signal Process. Lett., vol.
10, no. 10, pp. 289–292, 2003.
[3] X.P.V. Maldague, T.S. Jones, H. Kaplan, S. Marinetti and M. Prystay,
“Chapter 2: Fundamentals of Infrared and Thermal Testing: Part 1.
Principles of Infrared and Thermal Testing,” in Nondestructive
Handbook, Infrared and Thermal Testing, vol. 3, X. Maldague
a) b) technical ed., P. O. Moore ed., 3rd edition, Columbus, Ohio, ASNT
Press, 2001.
[4] V.A. Poryev, G.V. Poryev, “Experimental determination of the
temperature range of a television pyrometer,” in Journal of Optical
Technology, vol. 71, Issue 1, pp. 70-71, 2004.
[5] K. Pratt William, “Digital Image Processing,” John Wiley &Sons, 3rd
edition, 2001.
[6] A. Jain, “Fundamentals of Digital Image Processing,” Englewood
Cliffs, NJ: Prentice-Hall, 1989.
c) d) [7] J.A. Stark, “Adaptive image contrast enhancement using
generalizations of histogram equalization,” Image Processing, IEEE
Figure 6. Thermal Image Enhancement. Transactions on, vol. 9, no. 5, pp. 889-896, 2000.
[8] S.M. Pizer, E.P. Amburn, J.D. Austin, “Adaptive Histogram
The results with EME for different methods can be found Equalization and Its Variations,” Computer Vision, Graphics, and
Image Processing 39, vol. 355-368, 1987.
in Table 1.
[9] K. Zuiderveld, “Contrast Limited Adaptive Histogram Equalization,”
In: P. Heckbert: Graphics Gems IV, Academic Press, 1994
TABLE I. COMPARISON OF RESULTING EME’S OF DIFFERENT
ENHANCEMENT METHODS [10] Bichao Zhan, Yiquan Wu, “Infrared Image Enhancement based on
Wavelet Transformation and Retinex,” International Conference on
Images Original Histogram CLAHE Proposed Intelligent Human-Machine Systems and Cybernetics, pp. 313-316,
equalization method 2010.
Garden 10,92 27,33 16,47 47,50
[11] Chang-Jiang Zhang, Fan Yang, Xiao-Dong Wang, Hao-Ran Zhang,
Selma 3,82 23,20 10,68 43,45
“An Efficient Non-Linear Algorithmfor Contrast Enhancement Of
Street 5,48 11,87 9,94 23,40
Infrared Image,” International Conference on Machine Learning and
Cybernetics, pp. 4946-4951, 2005.
[12] R. Hummel, “Histogram modification techniques,” Comput. Graph.
IV. CONCLUSION Image Process., vol. 4, pp. 209–224, 1975.
[13] E. Wharton, S. Agaian, K. Panetta, “Comparative study of
We present novel enhancement technique based on a new logarithmic enhancement algorithms with performance measure,”
Proceedings of SPIE 6064, 342-353.
application of contrast limited adaptive histograms on
transform domain coefficients called logarithmic transform [14] Sos S. Agaian, B. Silver, K.A. Panetta, “Transform Coefficient
Histogram-Based Image Enhancement Algorithms Using Contrast
coefficient adaptive histogram equalization with local and Entropy,” Image Processing, IEEE Transactions, vol. 16, Issue 3, pp.
global processing. This strategy for image enhancement 741 – 758, 2007.
allows getting more contrast and detailed thermal image with [15] Sos. S. Agaian, “Visual morphology,” Proceedings of SPIE,
the following properties: irregular lighting and brightness Nonlinear Image Processing X, San Jose, CA, vol. 3646, pp. 139–
gradient. The proposed image enhancement results compare 150, 1999.
favourably against other state-of-the-art approaches. [16] Sos S. Agaian, Karen P. Lentz, and Artyom M. Grigoryan, “A New
Measure of Image Enhancement,” IASTED International Conference
ACKNOWLEDGMENT on Signal Processing & Communication, pp. 19-22, 2000.
[17] A. Silva Eric, Panetta Karen, Sos S. Agaian, “Quantifying image
This work was supported by Russian Ministry of similarity using measure of enhancement by entropy,” Proc. SPIE
Education and Science in frame of the Federal Program 6579, Mobile Multimedia/Image Processing for Military and Security
"Research and development on priority directions of Applications 2007, vol. 65790U, 2007.
scientific-technological complex of Russia Federation in [18] A. Berg and J. Ahlberg and M. Felsberg, “A Thermal Object
2014-2020" (contract № 14.577.21.0212 Tracking Benchmark,” Advanced Video and Signal Based
Surveillance (AVSS), 2015 12th IEEE International Conference on,
(RFMEFI57716X0212)). 2015.

Authorized licensed use limited to: Cornell University Library. Downloaded on September 02,2020 at 11:28:31 UTC from IEEE Xplore. Restrictions apply.