I.J.

Image, Graphics and Signal Processing, 2021, 6, 12-22

Published Online December 2021 in MECS ([Link]
DOI: 10.5815/ijigsp.2021.06.02

Image Recognition Using Machine Learning with

the Aid of MLR
Meherunnesa Tania
Department of Computer Science and Engineering, Jahangirnagar University, Dhaka, Bangladesh
Email: taniajucse25@[Link]

Diba Afroze
Department of Computer Science and Engineering, Jahangirnagar University, Dhaka, Bangladesh
Email: da.bony1971@[Link]

Jesmin Akhter
Institute of Information Technology, Jahangirnagar University, Dhaka, Bangladesh
Email: jesmin@[Link]

Abu Sayed Md. Mostafizur Rahaman

Department of Computer Science and Engineering, Jahangirnagar University, Dhaka, Bangladesh
Email: asmmr@[Link]

Md. Imdadul Islam

Department of Computer Science and Engineering, Jahangirnagar University, Dhaka, Bangladesh
Email: imdad@[Link]

Received: 10 August 2021; Revised: 02 September 2021; Accepted: 22 September 2021; Published: 08 December 2021

Abstract: In this paper, we use three machine learning techniques: Linear Discriminant Analysis (LDA) along different
Eigen vectors of an image, Fuzzy Inference System (FIS) and Fuzzy c-mean clustering (FCM) to recognize objects and
human face. Again, Fuzzy c-mean clustering is combined with multiple linear regression (MLR) to reduce the four-
dimensional variable into two dimensional variables to get the influence of all variables on the scatterplot. To keep the
outlier within narrow range, the MLR is again applied in logistic regression. Individual method is found suitable for
particular type of object recognition but does not reveal standard range of recognition for all types of objects. For
example, LDA along Eigen vector provides high accuracy of detection for human face recognition but very poor
performance is found against discrete objects like chair, butterfly etc. The FCM and FIS are found to provide moderate
result in all kinds of object detection but combination of three methods of the paper provide expected result with low
process time compared to deep leaning neural network.

Index Terms: Logistic regression, Eigen decomposition, objective function, scatterplot and entropy based combined
algorithm

1. Introduction

Nowadays object recognition/detection is widely used in the field of computer vision, robotics, medical image,
digital signal processing (DSP) etc. Object detection and recognition are quite close techniques, where object detection
is the technique of finding whether an object is present or not. Object detection is done by combination of image
processing and computer vision. In case of recognition, the object is not only detected but also identified as a type. The
main challenge of a researcher in object recognition/detection is to design an algorithm to acquire maximum possible
accuracy with minimum process time. Huge number of research works is found in recent literature for recognition of
objects and their modifications are also available to enhance the accuracy. Still no one is able to conclude that his or her
work is superior to another work. Here, we are going to review some of those works.
Human face recognition is widely used as a parameter in in the fields of biometric identification. The researchers
in [1] applied several machine learning techniques in detecting and then recognizing face with some adverse conditions
like, illumination, lighting, pose, unconstrained surroundings and so on. In the past years many approaches are already
have been discussed, but they are not well organized. Authors collect all these techniques or methods like Principal
Component Analysis (PCA), Artificial Neural Network (ANN), Support Vector Machine (SVM), Feature based

This work is open access and licensed under the Creative Commons CC BY 4.0 License. Volume 13 (2021), Issue 6
 Image Recognition Using Machine Learning with the Aid of MLR

approach, Gabor wavelet, GPU based face recognition to visualize or review those approaches in a systematic way. In
[2] authors match and recognize a scenario or a place by applying FuzzyArt. The proposed system is able to identify a
visual place taking both indoor and outdoor image as the input.
In [3], authors represented a machine learning approach for anticipation of short-term traffic system. The Linear
Discriminant Analysis (LDA) is used to predict real traffic, where 93% accuracy is claimed by the author. In [4] authors
made a comparison of efficiency between LDA and an artificial neural network in diagnosing breast cancer. Both
methods gave approximately 7-5% of false negative results but reducing some insignificant parameters and introducing
deep network architecture can minimize the error also visualized from the paper. Application of three common ML
algorithms: LDA, FCM and PCA are found in [5,6,7] in object recognition. These papers also showed the way of
enhancement of accuracy of object recognition, considering several ML techniques. Clustering of nodes in wireless
sensor network (WSN) is essential to reduce huge flooding of data, specially for large network case. Applications of
ML are also found in wireless communication system for signal detection. In [8] a new FCM based clustering algorithm
is proposed for WSN-based IoT applications. The outcome of the algorithm is to provide optimum clustering head and
minimum overall energy consumption. Finally, comparison is made with existing Low Energy Adaptive Clustering
Hierarchy (LEACH) protocol and found better result in contest of efficiency vs. round curve. An application of FIS is
found in [9], where detection accuracy of presence or absence of primary user (PU) at fusion center (FC) is determined
by FIS considering SNR, SIR and SINR as the input variables. The model used in the paper found closed results with
theoretical values.
Most of the previous works deal with object recognition using deep learning, which takes huge process time as
found in [10,11,12]. The main objective of the paper is to optimize accuracy rate and process time. Here we exclude
deep learning algorithm to save process time and choose three MLs as the combination of supervised and unsupervised
learning to enhance accuracy rate. In this paper we extract feature vectors of images then the dimension of data vector is
reduced using MLR. Finally, all three algorithms i.e. LDA, FIS and FCM are combined together in object recognition.
The main limitation of the paper is that accuracy of recognition is lower than the case of deep learning. Rest of the
paper is organized as: section 2 provides methodologies used in this paper to recognize images, section 3 deals with
results based on analysis of section 2 and section 4 concludes entire analysis with some future works.

2. Methodology

In this paper three ML algorithms: LDA, FIS and FCM with the help of MLR are used to recognize different types
of objects. Here class independent methods are joint together to cover wide variation of features of an image. The basic
theory and their combining scheme are given in this section.
2.1 Linear Discriminant Analysis (LDA)
There are two types of LDA technique used to deal with classification of object: class-dependent and class-
independent are discussed in [13]. Here, we only used class-independent method and their steps of operation is given
below:

1. Gi v e n th e d at a s e t, [x1 , x2 , x3 , . . . , xN ] u n d e r c cl a s se s [s1 , s2 , s3 , . . . , sc ].
2. Ca l c u l at e m e an (μj ) fo r j t h c la s s a n d m e a n (μ) of al l d a ta a s:

1 1 n
μj = ∑xi ∈sj xi a n d μ = ∑N c
i=1 xi = ∑i=1 μi
i
(1)
nj N N

3. Ca l c u l at e be t we e n- c l as s m at r ix a n d wi th i n -c l a ss m a tri x a s:
𝑛𝑛
𝐒𝐒𝑏𝑏 = ∑𝑐𝑐𝑖𝑖=1 𝑛𝑛𝑖𝑖 (μ𝑖𝑖 − μ) (μ𝑖𝑖 − μ)𝑇𝑇 a n d S w =∑𝑐𝑐𝑗𝑗=1 ∑𝑖𝑖=1
𝑗𝑗
( 𝑥𝑥𝑖𝑖𝑖𝑖 − μ𝑗𝑗 )(𝑥𝑥𝑖𝑖𝑖𝑖 − μ𝑗𝑗 )T, (2)

wh e r e x i j i s t h e i t h d at a p oi n t u n d er j t h cl a s s
−1
4. Ap p l y Ei ge n d ec o m p os it i o n o n Sw . Sb a n d s o rt t h e ei g e n v ec t o r s i n d e s ce n d i n g o r de r a c c o rd i n g t o t h e
e i g e n v a l u es .
5. Determine projection of data points of each class on the eigen vector of highest eigen value.
6. Make probability density function (pdf) of result of each class of training data under step 5.
7. Using pdf of step 5, we have to determine the class of test data set.

2.2 Multiple Linear Regression

The step of determination of linear equation under MLR can be describe as:

Volume 13 (2021), Issue 6 13

 Image Recognition Using Machine Learning with the Aid of MLR

1. Gi v e n th e d at a m at ri x i s X o f dim e n s io n of M×N a n d re s p o n s e v e c to r i s y o f M×1 .

2. De t e rm i n e t h e c o ef f ic i e nt ve c t or

, 𝛽𝛽 = (𝐗𝐗 ′ . 𝐗𝐗)−1 . 𝐗𝐗 ′ . 𝑦𝑦 = [𝛽𝛽0 𝛽𝛽1 𝛽𝛽2 . . . 𝛽𝛽𝑁𝑁−1 ] (3)

3. No w t h e e xp r e ss i o n o f M LR wil l b e ,

𝑦𝑦 = 𝛽𝛽0 + 𝛽𝛽1 𝑥𝑥1 + 𝛽𝛽2 𝑥𝑥2 +. . . +𝛽𝛽𝑁𝑁−1 𝑥𝑥𝑁𝑁−1 (4)

MLR is used in data classification algorithm to reduce dimension of data vector. Its application is found in big data
analysis like [14]. To keep the outlier within narrow range we also include logistic regression with MLR.
2.3 Fuzzy Inference System (FIS)
The working principle of the FIS consists of the following three steps like [15],

1. A F u z z i fi c a ti o n u ni t s u p p o rt s t h e a p pl i ca t io n o f n u m er o u s f u z z if i ca t i on me t h o ds , a nd
c o n v e r ts t h e cr i sp in p u t i n to f uz z y i n p ut .
2. Th e In f e r e nc e un i t c om b in e s s e v er a l F u z z y v ar i a bl e s ( ta k e n f r om t h e o u tp u t o f F u z z if i ca t io n
u n i t ) a n d p r o vi d e s F u z z y o u t p ut ba s e d o n F u zz y r u l e s.
3. Th e De f u z zi fi c a ti o n u n i t c o n ve r ts th e F u zz y o ut p u t of In f e r e nc e u ni t i nt o cr i s p o ut p u t.

2.4 Fuzzy c-means Clustering (FCM)

Fuzzy c-Means (FCM) is a data clustering technique in which a data set is grouped into c clusters with every data
point in the dataset belonging to every cluster to a certain degree discussed in [16]. The steps of FCM is given below.

Assume 𝑌𝑌 = {𝑦𝑦1, 𝑦𝑦2, . . . , 𝑦𝑦𝑛𝑛 } be the set of data points and 𝑉𝑉 = {𝑣𝑣1 , 𝑣𝑣2, . . . , 𝑣𝑣𝑐𝑐 } be the set of centers.

1. Ra n d o m l y s el e ct ‘c ’ a s t h e c l u st e r c e nt e r s.

2. Ca l c u l at e t h e f u z z y m e mb e r s hi p 𝜇𝜇𝑖𝑖𝑖𝑖 u s i n g:

1
𝜇𝜇𝑖𝑖𝑖𝑖 = 2 (5)
𝑑𝑑𝑖𝑖𝑖𝑖 𝑚𝑚−1
∑𝑐𝑐𝑘𝑘=1� �
𝑑𝑑𝑖𝑖𝑖𝑖

wh e r e , 𝜇𝜇𝑖𝑖𝑖𝑖 r ep r e se n t s t h e m e mb e r s hi p of i t h da t a t o j t h c lu s t e r c e nt e r.
c r e pr e s e nt s t h e n u mb e r o f c l u st e r c e nt e r.
𝑑𝑑𝑖𝑖𝑖𝑖 r e p re s e nt s t h e Euc l id e a n d i st a n c e b e t wee n i t h da t a a nd j t h c l us t er c e nt e r a n d m is
t h e f u z z i ne s s i n d e xe s .
3. Co m p u t e t he f uz z y c e n t er s v j u si n g :

∑𝑛𝑛 𝑚𝑚
𝑖𝑖=1(µ𝑖𝑖𝑖𝑖 ) 𝑥𝑥𝑖𝑖
𝑣𝑣𝑗𝑗 = , ∀𝑗𝑗 = 1,2, … 𝑐𝑐 (6 )
(∑𝑛𝑛
𝑖𝑖=1(µ𝑖𝑖𝑖𝑖 )
𝑚𝑚

wh e r e , v j re p r e se n t s t he j t h cl u s te r c e n te r a n d n i s t he n u m be r of da t a p o i nt s .
(𝑘𝑘+1)
4. Re p e a t st e p 2 a n d 3 u nt il �𝑣𝑣𝑗𝑗 − 𝑣𝑣𝑗𝑗𝑘𝑘 � < 𝛽𝛽 wh e r e, k is t h e i t er at i o n st e p a n d β i s t h e
t e rm i n at i o n c ri t er i o n

All the algorithms of this section will be combined to get better accuracy of recognition of images. The steps of
entire operation of image recognition is given below:

1. Read the image

2. Apply preprocessing (selection of region of interest, resizing and denoising) on the image.
3. Extract feature vector from the image.
4. Apply LDA, FIS, FCM and MLR + logistic regression on the vectors to recognize the category of image.
5. Determine probability of success, Psi in recognition of object against ith method.
6. Normalize the probability vector as, [𝑃𝑃𝑠𝑠1 𝑃𝑃𝑠𝑠2 𝑃𝑃𝑠𝑠3 ]/(𝑃𝑃𝑠𝑠1 + 𝑃𝑃𝑠𝑠2 + 𝑃𝑃𝑠𝑠3 ).
7. Determine entropy H of probability vector.

14 Volume 13 (2021), Issue 6

 Image Recognition Using Machine Learning with the Aid of MLR

8. If Psi is above the threshold τ and H is above the threshold ε then the object is considered to be recognized by
combined method.

Next section provides results based on above analysis.

3. Results and Analysis

This section provides results where our expectation is to get high accuracy of object recognition with minimum
possible process time. First of all, we consider images of human face from database Face-96 as the object against the
methodology of the paper. As the preprocessing, we eliminate the background from facial images of fig.1 using Viola-
Jones algorithm then we apply the resultant images as shown in fig.2 in LDA algorithm. The profile of LDA vector of
images along first four Eigen vectors are shown in fig. 3 (a)-(d), where the vectors of images from same person are
closed and vectors are found far apart for different person hence the LDA algorithm is suitable for human face
recognition like [17].

Fig.1. Facial images with background

Fig. 2. Elimination of background of facial images

Fourth Eigen Vector Third Eigen Vector

0 350
Person 1 image 1

Person 1 image 2
-50
300
Person 1 image 3

Person 1 image 4
-100
Person 2 image 1 250
Person 2 image 2

-150 Person 2 image 3

200
Person 2 image 4

-200

150

-250
Components along Eigen vector

Components along Eigen vector

100

-300

Person 1 image 1
50
Person 1 image 2
-350
Person 1 image 3

0 Person 1 image 4
-400
Person 2 image 1

Person 2 image 2
-50
-450
Person 2 image 3

Person 2 image 4

-500 -100

0 10 20 30 40 50 60 0 10 20 30 40 50 60

Index Index

(a) LDA vector along fourth Eigen value (b) LDA vector along third Eigen value

Volume 13 (2021), Issue 6 15

 Image Recognition Using Machine Learning with the Aid of MLR

Second Eigen Vector First Eigen Vector

0 -50

-50 -100

-100 -150

-150 -200

-200 -250

-250 -300
Components along Eigen vector

Components along Eigen vector

-300 -350
Person 1 image 1
Person 1 image 1
Person 1 image 2
Person 1 image 2
-350 -400 Person 1 image 3
Person 1 image 3
Person 1 image 4
Person 1 image 4
-400 -450 Person 2 image 1
Person 2 image 1
Person 2 image 2
Person 2 image 2
Person 2 image 3
-450 -500
Person 2 image 3
Person 2 image 4
Person 2 image 4

-500 -550

0 10 20 30 40 50 60 0 10 20 30 40 50 60

Index Index

(c) LDA vector along second Eigen value (d) LDA vector along first Eigen value
Fig.3. Profile of LDA vector of human face along first four eigen vector

Fig.4. Images of two types of objects

Fourth Eigen Vector Third Eigen Vector

1600 2000

Chair 1

1500 Chair 2

Chair 3

1400 Chair 4

Butterfly 1

Butterfly 2
1300
Butterfly 3 1500

Butterfly 4
1200

1100
Components along Eigen vector

Components along Eigen vector

1000
Chair 1
1000
Chair 2
900
Chair 3

Chair 4
800
Butterfly 1

Butterfly 2

700
Butterfly 3

Butterfly 4

600 500
0 10 20 30 40 50 60 0 10 20 30 40 50 60

Index Index

(a) LDA vector along fourth Eigen value (b) LDA vector along third Eigen value

First Eigen Vector

Second Eigen Vector
-600
2000
Chair 1

Chair 2
-800
Chair 3
1800
Chair 4

Butterfly 1
-1000
Butterfly 2
1600
Butterfly 3

-1200 Butterfly 4

1400

-1400
Components along Eigen vector
Components along Eigen vector

1200

-1600
Chair 1

Chair 2

1000
Chair 3
-1800
Chair 4

Butterfly 1

800 Butterfly 2
-2000
Butterfly 3

Butterfly 4

600 -2200

0 10 20 30 40 50 60 0 10 20 30 40 50 60

Index Index

(c) LDA vector along second Eigen value (d) LDA vector along first Eigen value
Fig.5. Profile LDA vector of chair and butterfly along first four eigen vector

16 Volume 13 (2021), Issue 6

 Image Recognition Using Machine Learning with the Aid of MLR

In human face detection the LDA vector of images along first four Eigen vectors work well but for other category
of objects (for example butterfly, chair, plane, car etc.) the model does not work properly. In this paper, we use a set of
images of chair and butterfly as shown in fig.4. The curves of LDA vectors are now found far apart against the objects
visualized from fig.5. In this case we include FIS of fig.6, where the output of LDA is taken as the input. We consider
the length of LDA vectors as 64×1, but only first six components of each vector (along first eigen vector) are shown in
Table 1. The upper part of Table 1 is for four images of chair, where ith column of the table correspond to first 6
elements of LDA vector of ith chair. Similarly, the lower part of the table is for four images of Butterfly. Each row of
the Table corresponds to input vector of particular type of image, which are used as the input of FIS. For example, from
the Table 1 the vector [2.9811 2.8033 3.0229 3.0235] is for chair and [3.0214 2.6609 3.0242 3.0242] is for butterfly.
We consider the output of FIS as 0 and 1 for chair and butterfly respectively also shown in Table 1. First of all, we
trained the FIS with 64 vectors then apply the test vectors for verification. The verification of Fuzzy rules for vector
[2.0979 2.3524 1.8622 1.7582] and [1.9648 1.6945 1.8004 1.8004] are shown in fig.7 and 8 respectively, where the
output is found very closed to 0 and 1.

Input -1

Input -2 Fuzzy
Fuzzy Output
Inference
System
Input -3

Input -4

Fig.6. FIS of four images under two types of object (4 chairs and 4 butterfly)

Table 1. LDA vector of images with FIS output

Chair as the object and LDA along 1st Eigen vector

FIS output
Image-1 Image-2 Image-3 Image-4

2.9811 2.8033 3.0229 3.0235 0

2.7106 2.5474 2.9957 2.8925 0

2.2744 2.4543 2.4682 2.2770 0

2.1184 2.3935 2.0840 1.8967 0

2.0979 2.3524 1.8622 1.7582 0

2.1270 2.3229 1.7366 1.7605 0

st
Butterfly as the object and LDA along 1 Eigen vector
FIS output
Image-1 Image-2 Image-3 Image-4

3.0214 2.6609 3.0242 3.0242 1

3.0220 2.3965 3.0242 3.0242 1

3.0207 2.4741 3.0242 3.0242 1

2.9721 2.4190 3.0238 3.0238 1

2.6019 2.3699 2.9619 2.9619 1

1.9648 1.6945 1.8004 1.8004 1

Volume 13 (2021), Issue 6 17

 Image Recognition Using Machine Learning with the Aid of MLR

Fig.7. Verification of Fuzzy rule with vector [2.0979 2.3524 1.8622 1.7582] to provide output of 0

Fig.8. Verification of Fuzzy rule with vector [1.9648 1.6945 1.8004 1.8004] to provide output of 1

(a) Image-1 vs. image-2 (b) Image-1 vs. image-3

18 Volume 13 (2021), Issue 6

 Image Recognition Using Machine Learning with the Aid of MLR

(c) Image-3 vs. image-2 (d) Image-2 vs. image-4

Fig.9. Surface plot of LDA vector of images

Finally, the surface plot of LDA vector of images are shown in fig. 9, where the transition from level 0 to 1 are
found gradual instead of sharp change of crispy logic. Therefore, there is a small possibility of error at the transition
border. Working on 100 different types of object with the length of Eigen vector of 64×1 we get accuracy of detection
of object as 89.34%.
Taking 22 row vectors of Chair and 22 row vectors of Butterfly (Table 1) we apply Fuzzy C-mean clustering
algorithm to segregate data of two different types of image. Considering columns corresponding to image 1 and 2 we
get the scatterplot of fig. 10(a). Again, considering columns corresponding to image 3 and 4 we get the scatterplot of fig.
10(b). Finally using MLR we convert four columns of Table 1 to two column then applying Fuzzy C-mean clustering
algorithm we get the scatterplot of fig. 10(c). The combination of MLR and Fuzzy C-mean clustering algorithm gives
better separation of points. It is worth to mention that MLR is combined with Logistic regression to keep data points
within the range of amplitude of [0 1] as shown in fig. 10(c).

3 3

2.8 2.8

2.6 2.6

2.4 2.4
Y

2.2 2.2

2 2

1.8 1.8
1.6 1.8 2 2.2 2.4 2.6 2.8 3 3.2 1.6 1.8 2 2.2 2.4 2.6 2.8 3 3.2

X X

(a) Image 1 and 2 (b) Image 3 and 4

0.9

0.8

0.7

0.6

0.5
Y

0.4

0.3

Class-1
0.2
Class-2
data1

0.1 data2

0
0 5 10 15 20 25 30 35 40 45
X

(c) MLR
Fig.10. Scattered plot of row vectors

Volume 13 (2021), Issue 6 19

 Image Recognition Using Machine Learning with the Aid of MLR

14

Image 1 and 2

Image 3 and 4
12 MLR

10

Obj. Function
6

1 2 3 4 5 6 7 8 9 10

Iteration Count

Fig.11. Convergence of objective function of Fuzzy C-mean clustering

The convergence of objective function of FCM from three cases: ‘Image 1 and 2’, ‘Image 3 and 4’ and MLR are
shown in fig.11, where convergence is found after 5 iteration for all the cases but variation of objective function is
found minimum for MLR case.

Table 2. Accuracy of Combined Method

Accuracy of detection
Type of
image Fuzzy C-mean Fuzzy C-mean Clustering +
LDA LDA+FIS LDA + FIS + FCN + MLR
clustering MLR
Human face 78.67% 85.42% 73.44% 77.10% 92.48%

Chair 83.21% 86.04% 74.37% 78.22% 87.67%

Butterfly 82.39% 84.11% 78.09% 78.34% 91.24%

Plane 79.06% 85.19% 72.66% 76.91% 90.12%

Fan 84.47% 87.08% 78.39% 78.58% 93.26%

We worked on images of different objects from benchmark database called Caltech 101,
([Link] and human face from another benchmark database called
Face-96. The Table 2 reveals the results of five types of objects. Here we combined the four algorithms LDA + FIS +
FCN + MLR using entropy based combining scheme of [18] as explained in previous section. The accuracy of
recognition of five types of objects are found different, as visualized from Table 2. Actually, accuracy depends on
quality of original image, size of input image (process time will be longer for larger size of image) and appropriate
preprocessing scheme of the image. The combined method gives accuracy of detection above 92% for all types of
object. We also measured the CPU time for five types of independent images of Table 2, where we got minimum CPU
time of 2.8 sec and maximum value of 3.12 sec. The target of the paper is fulfilled in a sense that moderate accuracy of
detection with minimum possible process time.

4. Conclusion and Future Works

In this paper we use three object recognition techniques: LDA, FIS and FCM, provided they are aided with MLR
give different results. Each technique has its own limitation, for example: LDA is appropriate for human face
recognition but has a poor performance for discrete objects. To enhance accuracy of recognition we combined LDA,
FIS, FCM and MLR and got the result above 92% for all types of object. In object recognition deep learning techniques,
which need big data set can enhances the accuracy of recognition till 98% at the expense of process time. The combined
model used in the paper is suitable for limited number of independent objects. In future we will include convolutional
neural network (CNN) to make a scenario of accuracy of recognition and process time as found in [19-20].

20 Volume 13 (2021), Issue 6

 Image Recognition Using Machine Learning with the Aid of MLR

References

[1] P. Borisagar, S. Jani, Y. Agrawal and R. Parekh, ‘An Efficient and Compact Review of Face Recognition Techniques,’ 2020
IEEE International Students' Conference on Electrical, Electronics and Computer Science (SCEECS), 2020, pp. 1-5, 22-23 Feb.
2020, Bhopal, India
[2] A. Dechemi and N. Achour, ‘An Approach of Data Fusion for FuzzyART Based Visual Recognition,’ 2019 6th International
Conference on Electrical and Electronics Engineering (ICEEE), 2019, pp. 86-90, 16-17 April 2019, Istanbul, Turkey
[3] D. Szostak, K. Walkowiak and A. Włodarczyk, ‘Short-Term Traffic Forecasting in Optical Network using Linear Discriminant
Analysis Machine Learning Classifier,’ 2020 22nd International Conference on Transparent Optical Networks (ICTON), 2020,
pp. 1-4, 22 September 2020, Bari, Italy
[4] G. B. G. Pereira, L. P. Fernandes, J. M. R. d. S. Neto, H. D. d. M. Braz and L. da Silva Sauer, ‘A comparative study of linear
discriminant analysis and an artificial neural network performance in breast cancer diagnosis,’ 2020 IEEE ANDESCON, 2020,
pp. 1-6, December 2020, Quito, Ecuador
[5] Tharwat Alaa, Gaber Tarek, Ibrahim Abdelhameed and Hassanien Aboul Ella, ‘Linear discriminant analysis: A detailed
tutorial,’ Ai Communications, vol.30, no.2, pp.169-190, 2017
[6] Y. Wang, K. Zhang and Z. Sun, ‘A Novel Deep-learning Pipeline for Light Field Image Based Material Recognition,’ 2020
25th International Conference on Pattern Recognition (ICPR), 2021, pp. 2422-2429, 10-15 Jan. 2021, Milan, Italy
[7] M. K. Halidu, P. Bagheri-Zadeh, A. Sheikh-Akbari and R. Behringer, ‘PCA in the context of Face Recognition with the Image
Enlargement Techniques,’ 2019 8th Mediterranean Conference on Embedded Computing (MECO), 2019, pp. 1-5, 0-14 June
2019, Budva, Montenegro
[8] Rahil Bensaid, Maymouna Ben Said and Hatem Boujemaa, ‘Fuzzy C-Means based Clustering Algorithm in WSNs for IoT
Applications,’ 2020 International Wireless Communications and Mobile Computing (IWCMC), pp.126-130, 15-19 June 2020,
Limassol, Cyprus
[9] Md Abul Kalam Azad, Anup Majumder, Jugal Krishna Das, Md Imdadul Islam, ‘Improving signal detection accuracy at FC of
a CRN using machine learning and fuzzy rules,’ Indonesian Journal of Electrical Engineering and Computer Science, vol. 21,
no. 2, pp. 1140-1150, February 2021
[10] E. Bayhan, Z. Ozkan, M. Namdar and A. Basgumus, ‘Deep Learning Based Object Detection and Recognition of Unmanned
Aerial Vehicles,’ 2021 3rd International Congress on Human-Computer Interaction, Optimization and Robotic Applications
(HORA), 2021, pp. 1-5, 11-13 June 2021, Ankara, Turkey
[11] S. Hayat, S. Kun, Z. Tengtao, Y. Yu, T. Tu and Y. Du, ‘A Deep Learning Framework Using Convolutional Neural Network for
Multi-Class Object Recognition,’ 2018 IEEE 3rd International Conference on Image, Vision and Computing (ICIVC), 2018, pp.
194-198, 27-29 June 2018, Chongqing, China
[12] L. Yan, Y. Wang, T. Song and Z. Yin, ‘An incremental intelligent object recognition system based on deep learning,’ 2017
Chinese Automation Congress (CAC), 2017, pp. 7135-7138, 20-22 Oct. 2017, Jinan, China
[13] Tharwat Alaa, Gaber Tarek, Ibrahim Abdelhameed and Hassanien Aboul Ella, ‘Linear discriminant analysis: A detailed
tutorial,’ Ai Communications, vol.30, no.2, pp.169-190, 2017
[14] X. Liu, Z. Tang, H. Huang, T. Zhang and B. Yang, ‘Multiple Learning for Regression in Big Data,’ 2019 18th IEEE
International Conference on Machine Learning and Applications (ICMLA), 2019, pp. 587-594, 16-19 Dec. 2019, Boca Raton,
FL, USA
[15] Pramit Dutta and Nafisa Anjum, ‘Optimization of Temperature and Relative Humidity in an Automatic Egg Incubator Using
Mamdani Fuzzy Inference System,’ 2021 2nd International Conference on Robotics, Electrical and Signal Processing
Techniques (ICREST), pp.12-16, 5-7 Jan. 2021, DHAKA, Bangladesh
[16] J. Sun, Y. Dai and K. Zhao, ‘DBFCM: a Density-Based Fuzzy C-Means with Self-Regulated Fuzzy Clustering Parameters,’
2020 39th Chinese Control Conference (CCC), 2020, pp. 2233-2238, 27-29 July 2020, Shenyang, China
[17] Md Abul Kalam Azad, Anup Majumder, Muhammad R. A. Khandaker, Jugal K. Das, Md. Imdadul Islam, ‘Primary User Aided
Cognitive Radio Network with Optimum Location of Relay,’ International Journal of Scientific & Technology Research,
Volume 9, Issue 09, pp.217-221, September 2020
[18] J. Yu and X. Lu, ‘Wafer Map Defect Detection and Recognition Using Joint Local and Nonlocal Linear Discriminant Analysis,’
in IEEE Transactions on Semiconductor Manufacturing, vol. 29, no. 1, pp. 33-43, Feb. 2016
[19] N. M. Tahir, Adam N. Ausat, Usman I. Bature, Kamal A. Abubakar and Ibrahim Gambo, “Off-line Handwritten Signature
Verification System: Artificial Neural Network Approach”, I.J. Intelligent Systems and Applications, vol.13, no.1, pp.45-57,
2021.
[20] Anozie Onyezewe, Armand F. Kana, Fatimah B. Abdullahi and Aminu O. Abdulsalami, “An Enhanced Adaptive k-Nearest
Neighbor Classifier Using Simulated Annealing”, I.J. Intelligent Systems and Applications, pp.34-44, vol.13, no.1, pp.45-57,
2021.

Volume 13 (2021), Issue 6 21

 Image Recognition Using Machine Learning with the Aid of MLR

Authors’ Profiles

Meherunnesa Tania has completed her [Link]. Engineering in Computer Science and Engineering from
Jahangirnagar University and pursuing her [Link]. from the same University. Her research interests are Machine
Learning, Robotics and Artificial Intelligence, Neural Network etc.

Diba Afroze has completed her [Link]. in Computer Science and Engineering from Jahangirnagar University,
Dhaka, Bangladesh in 2019. Currently, she is pursuing her [Link]. in Computer Science and Engineering from the
same institution. She is working in the area of Machine Learning. Her expanded Research interest is Deep
Learning, Natural Language Processing and Data Science.

Jesmin Akhter has received PhD degree in 2019 in the field of 4G wireless networks. from Department of
Computer Science and Engineering of Jahangirnagar University, Savar, Dhaka, Bangladesh and obtained [Link]
Engineering degree in Computer Science and Engineering from Jahangirnagar University, Savar, Dhaka,
Bangladesh in 2012. She also received her [Link]. Engineering degree in Computer Science and Engineering from
Jahangirnagar University, Savar, Dhaka, Bangladesh in 2004. Since 2008, she is a faculty member having current
Designation "Professor" at the Institute of Information Technology in Jahangirnagar University, Savar, Dhaka,
Bangladesh. Currently her research focuses are on IoT, network traffic, complexity and algorithms and software
engineering. Being a dynamic and versatile person who is capable of merging innovative ideas, technology, knowledge, and
experience for positive contribution towards the system development in the rapidly changing scenario of Information Technology and
become a good teacher in the field of software and telecommunication systems.

Abu Sayed Md. Mostafizur Rahaman has received PhD degree in 2014 from Department of Computer Science
and Engineering of Jahangirnagar University, Savar, Dhaka, Bangladesh and obtained [Link]. degree from Stuttgart
University at Stuttgart, Germany in Information Technology (INFOTECH) in the branch of Embedded System
Engineering in 2009. He received his [Link]. degree in Electronics and Computer Science, from Jahangirnagar
University, Savar, Dhaka, Bangladesh in 2003. Since 2004, he is a faculty member having current Designation
"Professor" in the Department of Computer Science and Engineering of Jahangirnagar University, Savar, Dhaka,
Bangladesh. During his graduation, he worked at BOSCH (biggest automobile company in Germany) as Trainee
engineer (Industrial internship) as part of his graduate degree in embedded Systems. Currently his research focuses on IoT, Digital
Forensics, Ethical Hacking, Web Security, Embedded Systems and S/W Systems.

Md. Imdadul Islam has completed his [Link]. Engineering in Electrical and Electronic Engineering from
Bangladesh University of Engineering and Technology, Dhaka in 1993 and [Link] Engineering from the same
university in 1998. He has completed CAS program from Asian Institute of Technology, Thailand in 2001. He has
done Ph.D. at the department of Computer Science and Engineering, J.U., Dhaka, in the field of Network Traffic.
He worked as an Assistant Engineer in Sheba Telecom (Pvt.) LTD (A joint venture company between
Bangladesh and Malaysia) from Sept'94 to July'96. He has very good field experiences in installation of Radio
Base Station and Switching Center for WLL. He is now working as a Professor, at the department of Computer
Science and Engineering, Jahangirnagar University, Savar, Dhaka, Bangladesh. His research field is network traffic, wireless
communications, wavelet transform, OFDMA, adaptive filter theory, neural network and machine learning. He has more than two
hundred research papers in national and international journals and conference proceedings.

How to cite this paper: Meherunnesa Tania, Diba Afroze, Jesmin Akhter, Abu Sayed Md. Mostafizur Rahaman, Md. Imdadul Islam,
" Image Recognition Using Machine Learning with the Aid of MLR", International Journal of Image, Graphics and Signal
Processing(IJIGSP), Vol.13, No.6, pp. 12-22, [Link]: 10.5815/ijigsp.2021.06.02

22 Volume 13 (2021), Issue 6