4th International Conference on Advanced Technologies MIA-32

For Signal and Image Processing ̢ ATSIPÿ 2018

March 21-24, 2018 ̢ Sousse, Tunisia

Histogram Equalization-Based Techniques for

Contrast Enhancement of MRI Brain Glioma Tumor
Images: Comparative Study
Hiba MZOUGHI 1 Ines Njeh1 Mohamed BEN SLIMA 1 Ahmed Ben Hamida 1
1ATMS Advanced Technologies for Medecine and Signal
ENIS, Sfax University, Tunisia
Email: [email protected]

are usually affected by noise and artifacts. In fact, the visual

Abstract— In Magnetic Resonance Imaging (MRI), the poor
quality of MR images has a direct impact on the accuracy of
images quality, particularly the artifacts inherent to this type of
images as well as the low contrast between tissues and inter- clinical diagnosis, which can be seriously limited by existing
individual variability, could make difficult the image analysis and noise arising during images acquisition. The major challenges
affect the accuracy of clinical diagnosis. Therefore, the needs for of preprocessing MRI images, besides the noise, are the low
image enhancement techniques arise to improve the relevant contrast between tissues and inter-individual variability, non
image contents through reducing the noise while preserving the intensity uniformity distribution and poor quality of acquired
actual details features. Various MRI images denoising techniques images. These limitations (drawbacks) could affects not only
have been proposed in literature where each technique has its the medical diagnostic tasks but also the ability of automatic
advantages and limitations. Among them, the Histogram computerized CAD tools for data analysis, such as the
modifications-based approaches arise as the most employed, by
segmentation and the classification of relevant features, the 3-
many researchers, for MRI contrast enhancement. This paper
presents a comparative study of the most histogram-based D images reconstruction and registration. Therefore, image
techniques, mainly AHE, CLAHE, BPDHE and AIR-AHE enhancement techniques are of great interest in MR imaging to
techniques, dealing with denoising and contrast enhancement improve the relevant image contents through reducing the
MRI images. Experimental study, using real-world databases, is noise while preserving the actual details features.
performed based on evaluation of quality measurement metrics: This paper is organized as follows; Section II provides a
absolute mean brightness error (AMBE), peak signal to noise summary of the related works dealing with the MRI contrast
ratio (PSNR) and Entropy. The studied histogram-based enhancement approaches. Section III presents a detailed
technique’s advantages and limitations are also discussed. description of the studied histogram-based techniques, mainly
Keywords—Contrast Enhancemet; Histogram Equalization; Adaptive Histogram Equalization (AHE) and Contrast-limited
Magnetic Resonance Imaging (MRI); Glioblastoma Brain tumor; Adaptive Histogram Equalization (CLAHE), Average
Intensity Replacement–Adaptive Histogram Equalization
I. INTRODUCTION (AIR-AHE) and Brightness preserving Dynamic Histogram
equalization (BPDHE). Section IV will provide experimental
MRI modality is a medical image technique relying on nuclear results, performed using real-world MRI brain databases, as
magnetic resonance (NMR) principals and produce images well as the performance evaluation of the studied techniques
illustrating various organs in the body. This technique using statistical metrics. Finally, Section V concludes the
captivates the signals resulting from the interaction between paper.
nuclear spins when submitted to an electromagnetic field [1].
The fields of exploration offered by MRI are wide: the II. RALATED WORKS
anatomical MRI is today the best suited non-invasive method
Over the past several years, numerous MRI contrast
to observe with a fine resolution the biological tissues, and
enhancement and denoising approaches have been reported in
thus particularly the brain tissues. Today, recent developments
the literature. This step of image preprocessing is very crucial
make it possible, with the same imager, to explore different
for the efficiency of the subsequent image processing steps,
aspects of the brain: brain activity with functional MRI
(fMRI), or connectivity of brain areas with diffusion MRI. for instance segmentation, classification and reconstruction.
This imaging modality has thus become an increasingly Among these approaches, one could report the linear filtering
central tool in brain medicine or in research into cognitive (spatial and temporal [2]) and nonlinear filtering approaches,
neuroscience. Hence, automatic interpretation of brain MRI e.g. anisotropic diffusion filter (ADF), bilateral filtering [3].
has become a major issue. Clinicians and cognitive scientists There are also the transform-based approaches, e.g. wavelet
need reliable tools to assist them in their decision- making and and curvelet transforms [4] as well as statistical-based
in the interpretation of the huge amount of created approaches e.g. maximum likelihood (ML), random forest and
information. Markov Random Field (MRF), etc.
MRI has boomed in recent years, and even the signal-to-noise For MRI contrast enhancement, the histogram modifications-
ratio and spatial resolution have been enhanced, MR images based approaches have gained a great research interest by

978-1-5386-5239-8/18/$31.00 ©2018 IEEE

scientists due to their advantages: simplicity and ease of use, preserve the best features of the classical HE technique in
higher SNR as well as its low computational complexity. In terms of brightness and contrast, while introducing
fact, the most employed classical images contrast modifications that will overcome the disadvantages associated
enhancement technique is the Histogram Equalization (HE), to HE. In the next subsections we are going to study and
which provides good performance for ordinary images, such describe some of these HE-based techniques.
as human portraits or natural images. However, in MRI brain
A. Adaptive histogram equalization (AHE)
images, there are often more complex situations where the
brain tumor White matter hyper intensities (WMH) signal is AHE is a local image processing technique developed by
high but it may mistakenly be considered as other brain tissues ZIMMERMAN et al. [8]. In this technique, for each pixel in
such as Cerebrospinal fluid (CSF). the image, a region centered by its pixel neighborhood, called
In this paper, we will focus on the study of the major ‘contextual region’, is assigned, as illustrated in Fig. 1. Then
histogram modifications based approaches when applied to histogram equalization mapping is applied to the pixels based
brain MRI images for contrast enhancement. Sazanita et al. [5] on the average intensity values in that region. Thus the
proposed a new approach applied for FLAIR MRI images, mapping applied for each pixel is different and it is adaptive to
known as the Average Intensity Replacement–Adaptive the local distribution of pixel intensities rather than the global
Histogram Equalization (AIR-AHE), which enhances the information content of the image [8].
contrast without over amplifying the entire image. It improves
the contrast of WMH relying on the intensity adjustment and
contrast mapping techniques, more details of this method are
explained in this paper. In [6], the authors proposed a N
Hierarchical Correlation Histogram Analysis algorithm N (x, y)
(HCHA) which provides an automatic contrast enhancement
of images during the examination of atrophic cell areas of
Parkinson Disease (PD) patients. This algorithm is based on
the grayscale distribution degree of pixel intensity by
constructing a correlation histogram. This method has as
objective to provide the best segmentation accuracy result and Fig. 1. The contextual region shown is a window of n X n pixels around a
facilitates subsequently CAD processes. pixel at location (x,y)
SENTHILKUMARAN et al. in [7] performed a comparative
Histogram mapping is performed using the following
study of different histogram-based techniques, mainly the
notations. If(x,y) is a pixel of intensity i from the image
histogram equalization (HE), BHE, modified BHE, AHE and
CLAHE, in order to enhance the contrast of general MRI brain
images. They used different evaluation metrics as Michelson ƒ m+− the mapping of right upper x+−
contrast, RMS contrast, absolute mean brightness error ƒ m++ the mapping of right lower x++
(AMBE) and Pixel Distance. ƒ m−+ the mapping of left upper x−+ and
ƒ m−− the mapping of left lower x−−
While the major comparative studies of histogram-based
techniques are performed on general MRI brain images, to the Next the neighboring regions are combined together using
best our knowledge, there is no comparative study performed bilinear interpolation in order to eliminate artificially induced
on MRI glioma brain tumors. In fact, the main objective of the boundaries to form the output enhanced image. The
present comparative study is to find, over the existing recent interpolated adaptive histogram equalization coefficient is
contrast-enhancement methods, the most preferment and computed according to the following equation [10]:
effective techniques for the MRI Glioma brain tumors images,
specifically glioblastoma. Since glioblastomas are infiltrative
tumors, their borders are often fuzzy and hard to distinguish
from healthy tissues [1]. Furthermore in some cases, the tumor where
region might share the same intensity profile with other
normal regions within the image [14].These characteristics (2)
make the localization of tumor core edges a very difficult task
AHE has the tendency to over amplify noise particularly in
for radiologists. Thus, the MRI contrast-enhancement is
homogeneous regions making an over-enhancement of the
important on the one hand, it is a clinical necessity that
image that creating objects which were not existing or visible
facilitate, for radiologists and neurologists, the tumor detection
on the original image. Contrast Limited Adaptive histogram
and distinguish it from other tissues (edema and necrosis), and
Equalization (CLAHE) attempts to overcome this limitation
on the other hand, it makes easier for the CAD to have more
by providing a clip limit factor.
efficiency and accuracy in different tasks such as automatic
segmentation and even the classification task. B. Contrast-limited adaptive histogram equalization
CLAHE differs from AHE in the ability to limit the over
III. HISTOGRAM EQUALIZATION BASED contrast enhancement with a predefined clip limit that depends
TECHNIQUES DESCRIPTION on the size of the neighborhood region and also on the
normalization of the histogram. Fig.2 presents the general
Contrast-enhancement is an essential pre-processing step
flowchart of CLAHE technique.
towards efficient clinical diagnosis of MR images. Various
HE-based contrast-enhancement techniques are reported in The nominal clip limit is calculated according to the following
literature. The main motivation of these variants HE is to equations [13]
 It is very difficult to find the exact location of local maximums
(3) due to histogram fluctuation without smoothing the histogram
and also the probability that there is some brightness levels are
Where NCLIP is the input value of clip limit and Navg represents missing. Thus the first step of BPDHE is to apply Gaussian
the average number of pixels expressed by: filter to the input histogram [14]. To keep the mean brightness
of the input image, BPDHE apply the brightness normalization
(4) process, as define by:
Where NGray is the number of gray level in contextual region, (5)
NCR-XP represents the number of pixels in X direction of
contextual region and NCR-YP represent the number of pixels Where Mi is the mean brightness of the input image, Mf is the
in Y direction of contextual region. CLAHE has a drawback mean brightness after the equalization process and f(x, y) is the
that it fails to retain the brightness with respect to the input output image after the equalization process.
image. Hence, to overcome this problem Brightness
Preserving Dynamic Histogram Equalization (BPDHE) has D. Average Intensity Replacement based on Adaptive
been considered in this paper. Histogram Equalization (AIR-AHE)
AIR-AHE is a recent contrast enhancement technique for brain
MRI images proposed by Iza et al. [5]. This technique,
Obtain Input image and input parameters composed of four processing steps, combines several
established enhancement algorithms:
Divide the input image into sub-images and Compute • Step1: Contrast stretching process where the
its histogram principal purpose is increasing the contrast of the
enhanced image.
Calculate the nominal clipping level from the actual • Step2: Apply the CLAHE technique with
clipping limits in the range of 0-1, in order to
Clipped histogram to the nominal clip level improve the contrast of the potential regions;
• Step3: Region of maximum intensity where
Interpolate gray level mapping the higher intensity location in the potential regions is
determined. It could be further improved by including
Output enhanced image the regional maxima function.
• Step4: Window sliding neighborhood
Fig. 2. Flow chart for CLAHE.
operation for intensity measurement. A sliding
neighborhood operation is used to determine the
C. Brightness Preserving Dynamic Histogram Equalization average intensity value of the pixel's neighborhood. It
(BPDHE)
is performed by sliding a set of pixels relative
Unlike other contrast enhancement techniques, BPDHE
divides the input image into sub-regions based on the position to the center of the pixel [4].
of the local maximums of input histogram. The next step is to
equalize each partition separately [9]. Before the equalized IV. EXPERIMENTAL RESULTS & DISSCUSSION
histogram taking place, BPDHE first map each sub image to a
new dynamic range. This change in the dynamic range will A. Data set
cause modification in the mean brightness of the output image, In order to perform the experimental results of the studied
thus the final step of this method is the normalization of the histogram-based techniques, several brain MRI images have
output intensity. Figure 3 describes the functional process of been used from the latest version of the Multimodal Brain
BPDHE. Tumor Image Segmentation database (BRATS 2015). This
database comprises around 300 subjects’ high- and low- grade
Input image glioma cases. Each subject contains T1 MRI, T1 contrast-
enhanced MRI, T2 MRI, and T2 FLAIR MRI volumes. The
Smooth the histogram with Gaussian
sizes of MRI images are 240 240 pixels. All datasets have
been aligned to the same anatomical template and interpolated
Detect the position of the local maximums from the to 1mm3 voxel resolutions. All the brain MRI images are co-
smoothed histogram registered and skull stripped.
The database is available via the Virtual Skeleton Database
Map each region into a new dynamic range. website (https://www.virtualskeleton.ch/) [11].

Equalize each partition independently B. Quality Measures

Quality measures are necessary to evaluate and to compare the
Normalize the image brightness
performance of the studied contrast enhancements techniques.
For this purpose, three quantitative evaluations metrics,
Fig. 3. Flow chart for BPDHE namely average absolute mean brightness error (AMBE), peak
signal-to-noise ratio (PSNR), and Entropy are used for
performance evaluation and whose mathematical expressions
are given as follows:
Absolute Mean Brightness Error (AMBE): It allows the
evaluation of the ability of preserving the image mean
brightness. It computes the difference of the mean brightness
between the original image and the enhanced one as it is
expressed by the following equation [12]:
(a) (b)
(6)
Where Iai and Iei represent respectively the mean intensity of
the whole acquired and enhanced images. Ideally, the mean
brightness of the enhanced image should be equal to the mean
brightness of the input image in which the value of AMBE is
zero. Thus a small AMBE is desired.

Peak signal-to-noise ratio (PSNR): It measures how much (c) (d)

the enhanced image has degraded when referred to the
acquired image.
(7)

Where Max (Iaq) 2 is the maximum intensity value of

the acquired image and Mean square error (MSE) is
the mean of intensities difference between acquired and (e) (f)
enhanced images, as it is expressed by the equation bellow. A
lower value for MSE means lesser error:

(8)

Entropy: It is applied to evaluate the amount of details

preservation in the enhanced image (higher entropy value is (g) (h)
desired). The entropy of the entire image is the summation of
all the individual entropies at every gray level defined. It is
given by the following equation [12][13]:

(11)

(i) (j)
C. Qualitative & Quantitative analysis
Two subject-cases were selected from the BRATS dataset for
performance evaluation of studied contrast enhancement
techniques. For each case, the best 10 slices have been chosen Fig. 4. Comparison of studied contrast enhancement techniques on Patient 2
of BRATS-2015 dataset: :(a) Original aquired image ; (b) histogram of
for further processing where the glioma tumor appears clearly. original aquired image ; (c) AHE result ; (d) histogram of AHE (e) CLAHE
The T1-weighted imaging, with contrast enhancement from result ; (f) histogram of CLAHE ; (g) AIR-AHE result ;(h) histogram of
gadolinium (T1Gd), has been chosen for preprocessing since it AIR-AHE ( i) BPDHE result; (j) histogram of BPDHE
is the best MRI modality to observe the hypo intense part of
the glioma tumor core. Figure 4 presents an example result The results shown in the Table I present the performance
using AHE, CLAHE, AIR-AHE and BPDHE techniques and evaluation of brightness preservation of various histogram-
their corresponding histogram. In general a good image based techniques studied in this paper.
enhancement should have a uniformly distributed histogram
without the presence of spikes
 TABLE I. COMPARISON RESULT OF STUDIED CONTRAST with a high grade Glioblastoma brain tumor from BRATS-
ENHANCEMENT TECHNIQUES FOR AMBE MEASURES
2015 database (figure7). Table IV, V and VI provide
Slice # AHE CLAHE AIR-AHE BPDHE respectively the analysis results for AMBE, PSNR and entropy
83 0.04 0.10 0.09 0.07 quality measures to compare the efficiency among the studied
84 0.04 0.10 0.10 0.07 contrast enhancement techniques.
85 0.04 0.10 0.09 0.07
86 0.04 0.10 0.09 0.07
87 0.05 0.11 0.09 0.08 TABLE IV. COMPARISON RESULT OF STUDIED CONTRAST ENHANCEMENT
88 0.04 0.10 0.09 0.07 TECHNIQUES FOR AMBE MEASURES TESTED ON PATIENT 07
89 0.03 0.09 0.09 0.06
90 0.02 0.08 0.10 0.05 Slice # AHE CLAHE AIR-AHE BPDHE
91 0.03 0.09 0.10 0.06 102 0.04 0.11 0.12 0.07
92 0.02 0.08 0.10 0.05 103 0.04 0.11 0.12 0.07
Average 0.03 0.10 0.09 0.06 104 0.04 0.12 0.12 0.07
(std) (0.008) (0.009) (0.003) (0.010) 105 0.04 0.12 0.12 0.06
106 0.04 0.12 0.13 0.08
Based on the observation of Table I, one could notice that 107 0.04 0.13 0.13 0.08
AHE seems to be the efficient technique in term of brightness 108 0.04 0.12 0.12 0.07
preservation followed by the BPDHE technique. Table II 109 0.04 0.12 0.12 0.07
presents the comparison results of Peak Signal to Noise Ratio 110 0.03 0.12 0.12 0.06
(PSNR) average values. 111 0.03 0.12 0.12 0.09
Average 0.04 0.12 0.12 0.07
TABLE II. COMPARISON RESULT OF STUDIED CONTRAST ENHANCEMENT (std) (0.002) (0.006) (0.004) ( 0.009)
TECHNIQUES FOR PSNR MEASURES

Slice # AHE CLAHE AIR-AHE BPDHE

83 20.51 17.17 18.20 17.05 TABLE V. COMPARISON RESULT OF STUDIED CONTRAST ENHANCEMENT
84 20.73 17.44 17.43 16.75 TECHNIQUES FOR PSNR TESTED ON PATIENT 07 OF BRATS
85 20.42 17.14 18.08 16.66
Slice # AHE CLAHE AIR-AHE BPDHE
86 20.34 17.09 18.11 16.37
19.80 16.34 18.41 15.49 102 22.32 16.78 16.54 16.26
87
88 20.68 17.13 18.31 16.17 103 22.04 16.21 16.18 15.56
89 21.40 17.78 18.22 18.03 104 22.11 15.87 16.19 15.75
90 22.90 18.78 17.82 19.28 105 22.29 15.64 16.19 16.65
91 22.65 18.24 18.13 17.85 106 22.02 15.18 15.83 14.68
92 22.91 18.47 18.19 18.72 107 21.62 14.58 15.23 14.58
Average 21.23 17.560 18.09 17.24 108 22.12 15.03 16.04 15.05
(std) (1.16) (0.71) (0.27) ( 1.19) 109 22.07 14.89 15.86 15.26
110 22.20 14.99 16.18 16.09
Quantitative analysis reveals that the AHE technique has 111 22.39 15.16 16.64 13.95
better performance in PSNR quality metric comparing to the Average 22.12 15.43 15.38 15.38
others contrast enhancement methods. Table III provides the (std) (0.21) (0.68) (0.84) (0.84)
comparison results between studied techniques for entropy
measurement.
TABLE VI. COMPARISON RESULT OF STUDIED CONTRAST ENHANCEMENT
TECHNIQUES FOR ENTROPY TESTED ON PATIENT 07 OF BRATS
TABLE III. COMPARISON RESULT OF STUDIED CONTRAST ENHANCEMENT
TECHNIQUES FOR ENTROPY MEASURES Slice # AHE CLAHE AIR-AHE BPDHE
Slice # AHE CLAHE AIR-AHE BPDHE 102 2.98 3.29 3.43 2.51
83 3.34 3.75 3.34 2.81 103 2.94 2.43 3.51 2.43
84 3.44 3.84 3.43 2.79 104 2.89 2.41 3.42 2.41
105 2.90 2.36 3.42 2.36
85 3.43 3.86 3.27 2.80
106 2.81 2.34 3.42 2.34
86 3.37 3.85 3.25 2.77
107 2.80 2.27 3.43 2.27
87 3.34 3.83 3.37 2.67
108 2.70 2.31 3.38 2.31
88 3.30 3.82 3.31 2.73
109 2.65 2.26 3.30 2.26
89 3.29 3.79 3.26 2.72
110 2.62 2.22 3.25 2.22
90 3.29 3.77 3.16 2.81 2.59 2.04 3.08 2.04
111
91 3.25 3.75 2.99 2.74 2.80 2.38 2.38
Average 3.36
92 3.25 3.74 2.83 2.66 (std) (0.14) (0.33) (0.84) ( 0.13)
Average 3,33 3.42 3.22 2,754
(std) (0.06) (0.04) ( 0.18) (0.05)
Also for this studied subject case of MRI glioma brain tumor,
AHE has better results in brightness preservation of the input
The CLAHE technique produces an enhanced image with the image and in PSNR quality metric than other histogram
highest entropy value, thus preserving more details in the equalization for contrast enhancement. However, AIR-AHE
image than the other compared techniques. technique provides better results in term of detail preservation.
The next subsection provides comparison results of the
studied contrast enhancement techniques tested on Patient 007
 method is evaluated relying on image quality measurement
mainly AMBE, PSNR and entropy. For the evaluation process,
we selected the most relevant slices where the tumor core
appears clearly, then we computed the average value (with a
standard deviation) of the quality evaluation metrics which
makes the evaluation to be more precise. Through this study
one could notice that, Adaptative Histogram Equalization
(AHE) technique provides efficient performances for MRI
(a) (b) contrast enhancement compared to other studied techniques.
For future works, we will focus on enhancing the AHE
techniques by introducing filtering approaches that could
improve the results in terms of accuracy and treatment
efficiency.

References
(c ) (d)
[1] HAVAEI, Mohammad, DAVY, Axel, WARDE-FARLEY, David, et
al. Brain tumor segmentation with deep neural networks. Medical image
analysis, 2017, vol. 35, p. 18-31
[2] MOHAN, J., KRISHNAVENI, V., et GUO, Yanhui. A survey on the
magnetic resonance image denoising methods. Biomedical Signal
Processing and Control, 2014, vol. 9, p. 56-69.
[3] ANAND, C. Shyam et SAHAMBI, Jyotinder S. Wavelet domain non-
linear filtering for MRI denoising. Magnetic Resonance Imaging, 2010,
vol. 28, no 6, p. 842-861.
(e) (f)
[4] ALEXANDER, M. E., BAUMGARTNER, R., SUMMERS, A. R., et
al. A wavelet-based method for improving signal-to-noise ratio and
contrast in MR images. Magnetic Resonance Imaging, 2000, vol. 18, no
2, p. 169-180.
[5] Isa IS, et al. Automatic contrast enhancement of brain MR images using
Average Intensity Replacement based on Adaptive Histogram
Equalization (AIR-AHE). Biocybern Biomed Eng (2017)
[6] Chiao-Min Chen1 et al , Automatic Contrast Enhancement of Brain MR
Images Using Hierarchical Correlation Histogram Analysis, J. Med.
Biol. Eng. (2015) 35:724–734 DOI 10.1007/s40846-015-0096-6
[7] SENTHILKUMARAN, N. et THIMMIARAJA, J. Histogram
equalization for image enhancement using MRI brain images. In
(g) (h) Computing and Communication Technologies (WCCCT), 2014 World
Congress on. IEEE, 2014. p. 80-83.
[8] ZIMMERMAN, John B., PIZER, Stephen M., STAAB, Edward V., et
al. An evaluation of the effectiveness of adaptive histogram equalization
for contrast enhancement. IEEE Transactions on Medical Imaging, 1988,
vol. 7, no 4, p. 304-312.
[9] IBRAHIM, Haidi et KONG, Nicholas Sia Pik. Brightness preserving
dynamic histogram equalization for image contrast enhancement. IEEE
Transactions on Consumer Electronics, 2007, vol. 53, no 4.
(i) (j) [10] KONG, Nicholas Sia Pik et IBRAHIM, Haidi. Color image
enhancement using brightness preserving dynamic histogram
equalization. IEEE Transactions on Consumer Electronics, 2008, vol. 54
[11] MENZE, Bjoern H., JAKAB, Andras, BAUER, Stefan, et al. The
Fig. 5. Comparison of studied contrast enhancement techniques on Patient 07 multimodal brain tumor image segmentation benchmark(BRATS). IEEE
from BRATS-2015 dataset: :(a) Original aquired image ; (b) histogram of transactions on medical imaging, 2015, vol. 34, no 10, p. 1993-2024
original aquired image ; (c) AHE result ; (d) histogram of AHE (e) CLAHE [12] TANG, Jing Rui et ISA, Nor Ashidi Mat. Adaptive image enhancement
result ; (f) histogram of CLAHE ; (g) AIR-AHE result ;(h) histogram of based on bi-histogram equalization with a clipping limit. Computers &
AIR-AHE ( i) BPDHE result; (j) histogram of BPDHE Electrical Engineering, 2014, vol. 40, no 8, p. 86-10.
[13] PATEL, Omprakash, MARAVI, Yogendra PS, et SHARMA, Sanjeev. A
comparative study of histogram equalization based image enhancement
V. CONCLUSION techniques for brightness preservation and contrast enhancement. arXiv
preprint arXiv:1311.4033, 2013
The main objective of this paper is to provide a comparative
[14] NJEH, Ines, AYED, Ismail Ben, et HAMIDA, Ahmed Ben. A
study of some existing techniques of contrast enhancement distribution-matching approach to MRI brain tumor segmentation. In
based on histogram equalization for MRI Glioblastoma brain : Biomedical Imaging (ISBI), 2012 9th IEEE International Symposium
tumor. Particularly the AHE, CLAHE, AIR-AHE and BPDHE on. IEEE, 2012. p. 1707-1710.
methods are explained and then compared. Each studied