A Review: Contrast-Limited Adaptive Histogram

Equalization (CLAHE) methods to help the

application of face recognition
Purnawarman Musa Farid Al Rafi Missa Lamsani
Faculty of Computer Science and Faculty of Computer Science and Faculty of Computer Science and
Information Technology Information Technology Information Technology
Universitas Gunadarma, Indonesia Universitas Gunadarma, Indonesia Universitas Gunadarma, Indonesia
Email: p musa@staff.gunadarma.ac.id Email: faridalrafi@gmail.com Email: missa@staff.gunadarma.ac.id

Abstract—Image processing technology in hardware and soft- Usually, the task of recognizing in computer science for face
ware has a significant development in all aspects of human recognition a someone based on identically from facial image.
civilization. Image processing is growing rapidly in various fields In the last decades, two method face recognition become very
such as medical image, Biometrics, Object identification, Object
character recognition (OCR), and many more. One biometric popular, the first is methods developed and the second are the
technique in an application detects and recognizes faces. Objec- high quality of the current videos/cameras.
tives of face recognition include recognizing emotions, identifying In the computer vision, the research of prototypes integrates
offenders, applying to security systems, image processing to ac- action with the result the retina analog functions (Smart Image
cess a door as an application of computer and human interaction. Sensor using analog Hmax) capable of realized autonomous
The facial recognition system in this research was conducted
live streaming using Local Binary Pattern Histogram (LBPH) vision systems accompanied by image treatments derived from
method. There is a constraint to a light intake received by the using 0.35µm CMOS technology as measurement and tools
camera through a sensor that captures the image. an image sensors. In the computer as autonomous vision
Based on the results of our research to make better facial systems, the action of mathematical calculation doesnt need a
recognition, we propose using adaptation methods on the intensity high precision and an analog filtering approach. Their project
of light that is around the image capture object. The method of
adaptation in this research is the Contrast Adaptive Limited proposal of prototypes integrates to embed a system on the
Histogram (CLAHE). retina analog functions into the image sensors. The superior
Index Terms—Image Processing, Biometric, Contrast-Limited position is projected in the instance of comparing with series
Adaptive Histogram (CLAHE), Local Binary Pattern Histogram of the digits 0 and 1 plan into effect is to specified condition
(LBPH), Face Recognition an approved of speed determination and power consumption
[2].
I. I NTRODUCTION Adaptive Histogram Equalization (AHE) tends to over am-
Now, in the world of information technology, biometrics plify noise in the relative region of an image. A variant
are relevant to the technology used for physical analysis and of Adaptive Histogram also Contrast-Limited Adaptive His-
human behavior for authentication. such as the introduction togram Equalization (CLAHE) aims to prevent over ampli-
of the physical human with the introduction of fingerprints, fication by limiting contrast amplification. The amplification
retina, iris, facial patterns, and signatures. contrast around the given pixel value is given by the slope
Face recognition support to human beings perform automat- function transformation. CLAHE limits present the amplifi-
ically any times, every day and practice with like easy task, cation by clipping based a predefined value in the histogram
but the process in the computer it has implemented to be a before CDF computing (cumulative distribution function) [3].
consisting of many different and connected parts of the task.
II. RELATED WORK
The computer needs dataset as variables that can the quality
of the methods, for example, version of illumination, low level A. Image Processing
of the resolution, occlusion, etc. The introduction of the future using digital image processing
The use of information technology in biometric tagging can as a basic concept and implementation for digital images is
be used in the identification model, where the biometric system the development phase of the field of study and the discipline
identifies each person and searches through the same code in of image processing, e.g: basic of images, compilation of
the database that has been compiled. The use of biometric is images in relating to spatial and frequency an area of image,
considered to have a higher level of security than traditional restoration, color unbalance, wavelets, image compression,
systems that use passwords or ID Card for authentication on morphology operation, segmentation image, motion and detec-
accessing a system [1]. tion objects, description of images, and the basics of objects
and patterns recognition [4]. • The variable L is the whole a number of gray levels
Expanded to development an image processing or novelty present at the image (usually 256),
in area image processing has coverage about deep learn- • The variable n is the number of pixels from dataset the
ing and deep neural networks, convolutional method, Neu- image, and
ral Networks, the Scale-Invariant Feature Transform (SIFT), • The variable Px (i) a representation of the image his-
Maximally-Stable Extremal Regions (MSERs), k-means clus- togram for the pixel into an index to i
tering and etc [5].
However, the image processing to an execution of these Algorithm 1 Histograms Equalization Function
rules to be followed in calculations is usually carried out in Require: ImageInput is ImageORIGIN AL , which ImageInput
digital form (using the computer) and need for a particular is convert to grayscale image
purpose intensive calculations not really compatible with the Ensure: ImageOutput is Image Histogram Equalized from
restriction of embedded systems. ImageInput
The embedded systems of camera sensors have the constraints 1: Using the function to read the image
of the quality of the images, a speed of treatment, the 2: Find the result into variable equ with convert image to
performance system of embedded processing, and low power histogram equalize
consumption [6]. 3: Stacking from the images side-by-side
4: Show the resulting image
B. Biometric
Before applied to computer technology, access systems on
a device still use special objects such as access cards or use When bringing with one quality image with distinct sources
memory power and the blocks are core low-level building of light dynamics conditions, the make the same in quantity
a multi-biometric system is implemented [7]. There are two and find to check the results.
reasons for using Biometric data today: Histogram equalization function has an approved of the
result during the histogram of the image outcome is confined
1) As a verification tools, where the required data is on an
to an intense region. It inoperable places where the relatively
object (such as a document or credit card). But they can
great size intensity difference and histogram covering an area,
be duplicated so that it can be misused by someone else.
which used to refer to two reflecting a lot of light and no light
2) As a verification codes, usually using a password. Some-
as representation gray level pixels are present in figure 1.
times we forget the codes. Even if we use the best
encryption algorithm, there are still algorithms that can
do the decryption
The use of biometrics has several advantages, as follows:
• Can not be lost or forgotten
• Difficult to duplicate, share or change hands
• Authenticity is guaranteed

C. Contrast-Limited Adaptive Histogram Equalization

(CLAHE)
Contrast-Limited Adaptive Histogram Equalization is an
image processing technique been used to make greater in the
contrast of an image. CLAHE operates in a small area in an (a) Original Image without (b) Result Image with AHE
image, known as a tile call, rather than an entire image. The AHE
Fig. 1. Result image use AHE method
application of the histogram is separate from the histogram
in general, where the adaptive method is to calculate the The histogram equalization calculates the global contrast
function of the state of being contrast color to change every from the image. For a few cases, the image show unclearness
tile. Each tile contrast increases the histogram of the result result. For example, this images shows an input image (in
on region approximation to it by the distributed value. The figure 2(a)) and the figure 2(b) is result after global histogram
neighboring tiles then merged to avoid induced boundaries. equalization.
The contrast, mainly in homogeneous areas, will be limited to The scenery behind the main object of contrast has become
avoid amplifying a noise that could be present in the image. better after the result of the image for histogram equalization.
In figure 1, the Histogram Equalization is gray on the image The form the comparative and superlative of the face of the
x and the value i is a gray level amount to a maximum value statue in both images (between in fig. 2(a) and fig. 2(b)). Raw
of 256. Whereas to look for Probralities at pixel image from data for image is lost of the information as an image, if due
level i in an image to calculated as follows; to over-brightness.
ni From this problem, adaptive histogram equalization is used.
Px (i) = P(x=i) = , ⇒ 0 ≤ i < L (1)
n In this section, the image is divided into small blocks called
which, tiles (tile size 8x8). This tiles used to refer to every one of
Algorithm 2 Contrast-Limited Adaptive Histogram Equaliza- algorithm in facial recognition was developed by Sirovich and
tion (CLAHE) Function Kirby (1987) and published by Matthew Turk and Alex Pent-
Require: ImageInput is ImageORIGIN AL land in face classification (Turk, Pentland 2002). Classification
Ensure: ImageOutput is Image Contrast-Limited Adaptive is obtained by comparing face data from existing face database
Histogram Equalization (CLAHE) [8].
1: Find and read the image to put into variable img b) Fisherfaces: is a facial recognition algorithm devel-
2: Call function of CLAHE with clipLimit=2.0 and matric = oped using Principal Component Analysis (PCA) which is
8x8 also the core of the Eigenface method, and is found by
3: The result CLAHE into variable cl1 statistician Sir R. A. Fisher (Fisher, 1936). The Fisherfaces
4: Show the result CLAHE image method uses the class-specific transformation matrix, so they
do not capture illumination as obviously as the Eigenfaces
method. It’s important to mention, that the performance of the
Fisherfaces heavily depends on the input data as well [9].
c) Local Binary Pattern Histogram (LBPH): is a facial
recognition technique with a method of extracting local fea-
tures from images [10]. LBPH was first introduced in 1994.
Since it was discovered, it has become a good technique in
terms of texture classification.
The spatially enhanced Local Binary Pattern (LBP) and
Local Binary Pattern Histogram (LBPH) a system of methods
used in a particular area has succeed in achieving an estab-
lished position in the field of Face Recognition (FR) and obtain
(a) Original Image as image test something from many face analysis approaches [11].

E. LBPH in face recognition

Local Binary Patern Histogram (LBPH) is one of the applied
face recognition using binary pattern. This algorithm looks
like a very simple but very effective and efficient way to
describe the texture and shape of a digital image. LBPH has
simpler computing, making it a popular approach in various
applications [12].
The LBPH method is operated by dividing an image into
(b) Result Image using AHE (c) Result Image using a small regional. The introduction of the operator’s texture to
CLAHE the pixel label of a image with the environment of each pixel
Fig. 2. CLAHE method testing
and represents the result on a number binary.
The concept of using LBP is simply described as follows:
two or more of these blocks is histogram equalized as usual. 1) Divide the image to be used into multiple matrix (ex-
The results that in a small area becomes the histogram would ample divided by 16 matrices)
be limited to a small region. If the noise is there, it will 2) Each pixel present in a matrix will be compared with
be amplified. To render avoid the noise, contrast limiting is pixels from the 8 neighbors (top left, middle left, lower
relevant. If any histogram bins are above the specified contrast left, right top, and so on).
limit, those pixels are cut (clipped) and distributed uniformly 3) If the middle value is greater than the pixel value then
to other bins before applying histogram equalization. After give the value ”0” and if the middle value is smaller
equalization, to remove artifacts in tile borders, bilinear inter- than the pixel value it will give value ”1”. Then we will
polation is applied. See the result of the image (in figure 2) get an 8 digit value of binary value.
and compare it with results of image (in figure 2(c)), especially 4) Calculate the value of histogram, matrix, frequency of
the statue region. each number obtained (ie, any combination of which
is smaller and larger than the center. The value of this
D. Face Recognition Algorithm histogram is obtained in 256 vector dimensions.
The face recognition method is used to get geometric 5) Using Normalization function at histogram value.
features of a face to recognize a human face. The most famous 6) Sync the histogram value of each matrix, so it will give
algorithm is 3 (three) Face recognition algorithms that can be the vector feature value of the whole image.
used include: 7) In face recognition, the captured value will be compared
a) Eigenfaces: is one model of face recognition algo- to the histogram value, if close to the same ideal value,
rithm that uses eigenvector. The inventor of the eigenface then the face will be recognized as the user.
 III. METHODS AND IMPLEMENTATION Algorithm 3 Contrast-Limited Adaptive Histogram Equaliza-
tion (CLAHE)
In the systematic investigation into and study our project,
we discuss an image contrast enhancement algorithm speci- Require: Read the indexed color image as Original Image
fying the image overcomes limitations in standard Histogram Ensure: Showing the image result into the RGB color
1: Cause to change in form the listing image in an index
Equalization (HE). In preference to implementing our research
for contrast enhancement, there are two primary features. The (each pixel) expressing an act a true-color (RGB) image
2: Become different the RGB image using function change
first feature is called Adaptive Histogram Equalization (AHE),
which a wide divergence between two groups the images into the L*a*b* color space
3: Specified scales from values to the area of variation be-
into regions and performs local HE, and the second feature
is a specified name Contrast-Limited Histogram Equalization tween upper and lower limits on expected by the Adaptive
(CLAHE), which reduces noise by partially reducing the local Histogram Equalization (AHE) function
4: Present CLAHE method on the L a part of channel using
Histogram Equalization. Bilinear interpolation is used to avoid
visibility of region boundaries. Number of tiles and Contrast enhancement limit
5: Plotting the scale the result to return on the range the state
A. Experiment of Contrast enhancement limit of being used for purpose L*a*b* color space
Introducing the Contrast enhancement limit, an activity 6: To change the following as the consequence image back
identifies clearly and definitely like actually existing a scalar into the RGB color space
quantity in the range [0, 1]. The resulting values are higher
than the limit, then we need to add more contrast. The function
Contrast Limit features a contrast that contributes to a result In this research, we conducted experiments in same places,
that keeps oversaturation of the image. These areas are typical but different illumination to prove the application of CLAHE
by a high peak in the histogram of the individual item image obtained clear image results in dark conditions or bright
tile due to many pixels grading in the same gray level (see conditions, as shown in Figure 4.
fig. 3).

(a) ClipLimit=0.5

(a) Dark conditions (b) Bright conditions

Fig. 4. The result of Image using CLAHE method with different illumination

C. Experiment of Local Binary Patterns Histogram approach

for Face Recognation
The implementation of LBPH by every one of an image
must be analyzed to doing independently. The LBPH method
(b) ClipLimit=1.0 (c) ClipLimit=2.0 is a moderate extent simpler to analyze, in the perceived by a
sense that characterizes each image in the dataset locally; and
Fig. 3. Compare of contrast enhancement limit
on which a newly cannot describe and provide as an image,
Without the clipped limit being the case that, it could things the LBPH is better method analysis on face recognize and
that have been the result in some cases it less good the original present the different from the result to each of the datasets
image. images. This method has provided analyze to the images the
local patterns in each position in the image.
B. Experiment of CLAHE method with illumination condition For detecting a face from an image, the method will return
Based on the CLAHE algorithm shown in alg. 3, then the x and y coordinates of used to refer to the whole a faces in
the first CLAHE function Read the indexed color image as a dataset. The process of thinking, that application can detect
Original Image. After the CLAHE function is executed, it will only facial recognize. A human being regarded as the tester
get as shown in figure 4. a recognizer has the definite about face recognition, using the
LBPH method to recognize a specific person will be given a in twins can detect and recognize their faces. The author has
vector of faces belong to the person. obtained permission from the people in the picture.
For this, We propose the following algorithm 4:
D. Experiment of Detection and Recognation with Two Face
Algorithm 4 LPBH for Face Recognation per Frame
1: Read the input video for the next frame From the test result in table II, this experiment test intend
2: The clasification the faces from detect all in frame to detect and recognize every face in one frame. The first user
3: Recognize each faces for the detected faces was tested whether it can detect and recognize the face of the
4: The actions are successful, produce a green rectangle user. Then, testing on the second user was tested to detect
5: Write the accuration with the level of each face in the and recognize the face of a second user. The final test of this
frame to a window experiment is testing by capturing images that contain more
6: Repeat steps 1-5, until no more frames in the input video than one face in the image frame. From the results in the 3rd
column in table II, the face detection and facial recognition
In instruction to test the application using the training set experiment for one frame is more than one face successfully
from dataset images. For example, looking to analysis, and recognized and applied to our research.
that the results have discerned visually illustrating a general
of detection process and process of recognition of faces. The TABLE II
theory the back side is quite interesting (see table I). D ETECTION AND R ECOGNATION T WO FACE FOR O NE F RAME
Testing User 1 Testing User 2 Testing Both Users

TABLE I
I MPLEMENTATION OF CLAHE FOR FACE R ECOGNATION
Case Face Recognation Description of the Image

Capture Image:
1. Exterior Room
1 2. Lumen = 240 lx
3. Accuration Recognation = 75%
4. Face Detect founded
5. Face Recogntion founded

Capture Image:
1. Exterior Room
2 2. Lumen = 198 lx
3. Accuration Recognation = 85%
4. Face Detect founded
5. Face Recogntion founded

Capture Image:
1. Interior Room
3 2. Lumen = 98 lx
3. Accuration Recognation = 95%
4. Face Detect founded IV. CONCLUSION
5. Face Recogntion founded
From the improved image results based on the contrast
Capture Image: boundary that has been adapted HE, it can improve the image
1. Interior Room obtained. If the contrast that captured by its image is very
4 2. Lumen = 65 lx
3. Accuration Recognation = 50%
dark, the contrast value will be brightening on the image
4. Face Detect founded result. Conversely, if the image captured is very bright, then
5. Face Recogntion founded the contrast value will be reducing on the image result and
Capture Image:
become darker.
1. Interior Room The application of LBPH method for face recognition on
5 2. Lumen = 65 lx this test results have been done, where this research can show
3. Accuration Recognation = 95%
4. Face Detect founded
detection and face recognition successfully recognize the face
5. Face Recogntion founded of every user who is outside and indoor received through
the camera. But when testing twins, sometimes it resulting
Experiments in the image results of the CLAHE test above error when doing face detection and recognition incorrect face
in table I, there is outdoor testing and indoors that result in matching.
different lumens. Each the result can recognize the face of the The continuation of this research for appropriate future
user tester, where every previous face has been stored in a technologies, as biometric data for the authentication of a
database. There are also tests on twins that have been tested system. Examples of applications that can be done are the
in cases 4 and 5. Although the success to detect and recognize attendance system and automatic door access by using a
the face of one of the twins only succeeded 50%, the results recognizable user face.
 V. ACKNOWLEDGEMENT
This research was fully given assistance to realize this
project by the Department of Computer Science. The authors
thankfully acknowledge Universitas Gunadarma for especially
financially make available for use research funding and autho-
rization to access laboratory in using the research facilities.
R EFERENCES
[1] P. Musa, N. Yuliani, and M. Lamsani, “Rancang bangun pengendali
pintu automatis berdasarkan ciri wajah menggunakan metode euclidean
distance dan fuzzy c-mean,” 2008.
[2] M. Paindavoine, J. Dubois, and P. Musa, “Neuro-inspired smart image
sensor: analog hmax implementation,” vol. 9403, 2015, pp. 9403 –
9403 – 8. [Online]. Available: https://doi.org/10.1117/12.2079093
[3] S. M. Pizer, E. P. Amburn, J. D. Austin, R. Cromartie, A. Geselowitz,
T. Greer, B. ter Haar Romeny, J. B. Zimmerman, and K. Zuiderveld,
“Adaptive histogram equalization and its variations,” Computer Vision,
Graphics, and Image Processing, vol. 39, no. 3, pp. 355 – 368, 1987.
[4] A. Rosebrock, Practical Python and OpenCV: An Introductory, Example
Driven Guide to Image Processing and Computer Vision. PyImage-
Search, 2014.
[5] R. C. Gonzalez and R. E. Woods, Digital image processing, 4th ed.
Upper Saddle River, N.J.: Prentice Hall.
[6] P. Musa, S. A. Sudiro, E. P. Wibowo, S. Harmanto, and M. Paindavoine,
“Design and implementation of non-linear image processing functions
for CMOS image sensor,” in Optoelectronic Imaging and Multimedia
Technology II, vol. 8558, 2012, p. 85580O.
[7] R. I. Hammoud, B. R. Abidi, and M. A. Abidi, Face Biometrics for
Personal Identification: Multi-Sensory Multi-Modal Systems. Springer,
2007, p. 275.
[8] A. L. Ramadhani, P. Musa, and E. P. Wibowo, “Human face recognition
application using pca and eigenface approach,” in 2017 Second Interna-
tional Conference on Informatics and Computing (ICIC), 1-3, pp. 1–5.
[9] P. N. Belhumeur, J. P. Hespanha, and D. J. Kriegman, “Eigenfaces vs.
fisherfaces: Recognition using class specific linear projection,” IEEE
Trans. Pattern Anal. Mach. Intell., vol. 19, no. 7, pp. 711–720, Jul.
1997.
[10] H. Dienillah, “Aplikasi pengenalan wajah menggunakan algoritma lbph
berbasis opencv,” 2014.
[11] B. Yang and S. Chen, “A comparative study on local binary pattern
(lbp) based face recognition: Lbp histogram versus lbp image,” Neuro-
computing, vol. 120, pp. 365 – 379, 2013, image Feature Detection and
Description.
[12] K. Salton, “Face recognition: Understanding lbph algorithm,” Towards
Data Science, 2017.