Professional Documents
Culture Documents
Article QR
Author (s): Isra Anwar, Syed Farooq Ali, Jameel Ahmad, Sumaira Zafar, Malik Tahir Hassan
DOI: https://doi.org/10.32350.icr.22.05
History: Received: September 9, 2022, Revised: November 29, 2022, Accepted: December 10, 2022
A publication of
School of Systems and Technology
University of Management and Technology, Lahore, Pakistan
Feature Based Techniques for Face Recognition using
Supervised Learning Algorithms Based on Fixed
Monocular Video Camera
Isra Anwar1, Syed Farooq Ali 2, Jameel Ahmad3, Sumaira Zafar4*, and
Malik Tahir Hassan5
School of Systems and Technology, University of Management and Technology,
Lahore, Pakistan
Abstract- Automatic face recognition depicted that the proposed
has ample significance in biometric approaches achieve a percentage
research. Recent decades have accuracy of up to 95.5%.
witnessed enormous growth in this
Index Terms- face identification, face
research area. Face-based
recognition, geometric features,
identification is always considered
supervised learning, Support Vector
more expedient as compared to other
Machine (SVM)
biometric authentications owing to
its uniqueness and wide acceptance. I. Introduction
The major contribution of this work
is twofold; firstly, it comprises an Despite the fact that numerous
extension of manual thresholding algorithms have been proposed in
feature-based face recognition the literature, face recognition in
approach to an automatic feature- unconstrained scenes remains a
based supervised learning face challenging problem due to changes
recognition. Secondly, various new in appearance caused by the
feature sets are proposed and tested variations in pose and expression.
on several classifiers for 2, 3, 4, and 5 Still, face recognition has achieved
persons. In addition, the use of slope mammoth popularity during the last
features of facial components, such few decades in computer vision,
as the nose, right eye, left eye, and
pattern recognition, neural
lips along with other conventional
features for face recognition is also a networks, neuroscience, cognitive
unique contribution of this research. science, image processing,
Multiple experiments were psychology, and physiology for
performed on the UMT face good reasons.
database. The results demonstrated a The major reasons for its wide
comparison of 5 different sets of
acceptance are its simplicity,
feature-based approaches on 7
classifiers using the metrics of time security, and authenticity. Face
efficiency and accuracy. They also authentication works regardless of
*
Corresponding Author: sumairazafar.sz@gmail.com
School of System and Technology
83
Volume 2 Issue 2, Fall 2022
Feature Based Techniques for...
any user input in the form of features are suggested for feature
passwords, PIN codes, plastic cards, matching. Thirdly, a detailed
keys, tokens, and smart card comparison of these approaches is
authentication, which makes it provided using 7 existing classifiers
simple. As far as security is on a dataset of 2, 3, 4, and 5 persons.
concerned, all other methods which The proposed work uses the
involve user input are no match for slopes of different fiducial
face recognition, as user input in any landmarks of facial components
aforementioned form always comes (nose, right eye, left eye, and lips) to
with the risk of its misuse. This is build a slope table that is merged
exemplified by the use of passwords with other geometric and holistic
and PINs, which can be guessed or features. Finally, features are fed to
stolen. the classifiers for training and
Moreover, the solicitude of matching. The rest of the paper is
enhanced physical security systems organized as follows. In Section 2, a
has enjoyed rapid growth in recent review of the related literature is
years due to a dramatic increase in presented. Section 3 gives an
the crime rate. Face recognition overview of the data set used for
plays a pivotal role in this regard. It experiments. It also incorporates the
is widely used in numerous proposed framework. Experimental
applications, such as automatic evaluation is presented in Section 4.
attendance systems, physical access Finally, conclusions are drawn in
control, surveillance control, people Section 5.
tagging, gaming, image search, and
II. Literature Review
many more. This work proposes
several feature-based approaches of Face recognition has been an
face recognition containing some active area of research for the last
new features, namely slope table, few decades. Initially, a semi-
along with random projection and automated system was proposed for
regional properties. face recognition and only major
features, such as ears, eyes, mouth,
There are three major and nose were used for this purpose
contributions of the current study. [2]. The algorithms of face
Firstly, it proposes a new approach
recognition can be divided into two
for face recognition which extends
broad categories, namely feature-
the manual thresholding system [1]
based techniques and holistic
to a supervised learning algorithm. approaches [3]. Feature-based
Secondly, five different sets of new approaches initially identify and
Innovative Computing Review
84
Volume 2 Issue 2, Fall 2022
Anwar et al.
1
https://sites.google.com/site/farooq1us/dat
aset
Innovative Computing Review
90
Volume 2 Issue 2, Fall 2022
Anwar et al.
Fig 4. Facial fiducial points are detected and shown on the facial image at
the right side by applying BoRMaN on the left image
B. Features Reckoning collection, whereas the fifth
approach is a combination of
Facial fiducial points computed
Approach 4 in combination with the
in the previous phase were used to
slope table.
calculate different features, using
several approaches given to the 1) Approach 1: Random
classifiers for training and testing. projection features (RPF)
Identification labels of these This is a holistic feature
features were passed as input to a extraction approach that is time
classifier for training and testing efficient and based on features
purposes. Face recognition is based extracted from the facial image and
on the output of the classifiers. If the sparse matrix. Initially, features are
values of the features determine the extracted effectively using a sparse
difference as below the threshold measurement matrix and then
during the testing phase, then as per multiplied with the image features.
the matching frame, the classifiers This model employs random
are automatically classified. The projection which preserves the
proposed five different methods to structure of an image. A random
extract features from 2D frames are
matrix Z 𝜖𝜖𝑍𝑍 (n∗m) whose rows have
explained below. The first four
unit length projects data from the
techniques differ by feature
Innovative Computing Review
94
Volume 2 Issue 2, Fall 2022
Anwar et al.
Fig. 5. Explaining random projection that converts image features into the
compressed vector using sparse measurement matrix
In the next step, a histogram of frequency. The histogram of
this magnitude is calculated to magnitude is also considered in
divide the entire range of values into stating distinctive values for each
a series of intervals. This reduces image. The overall process of
the feature vector size based on feature calculation using a
The major and minor length axis ellipse that has the same second
specifies the length of the major and moments as the region. The solidity
minor axes of the ellipse that have of the region is calculated by the
the same normalized second central proportion of pixels in the convex
moments as the region, respectively. hull, as shown below in Eq. 9.
Orientation states the angle between 𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴
𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 = (9)
the x-axis and the major axis of the 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶
The ratio of length to width of using points 1 (x1; y1), 2 (x2; y2) and
the left eye in Fig. 8(a) is computed 3 (x3; y3), 4 (x4; y4), respectively.
with eye length and width. Eye The ratio of length to width of the
length and width are calculated by right eye is also computed in a
the Euclidean distance formula similar manner.
However, this approach also uses In the above equation, (𝑋𝑋𝑖𝑖 , 𝑌𝑌𝑖𝑖 )
the slope features of a facial image are the two coordinates of the fifth
along with the ratios of length to point of a facial component. 𝑃𝑃𝑥𝑥 and
width, area, and angles. Ratios and 𝑃𝑃𝑦𝑦 are Euclidean distances of the
angles are computed using Eqs. [13- boundary point to the centroid along
18]. For slope features, the proposed the x-axis and y-axis, respectively. P
framework introduces a represents the line from the
groundbreaking technique, where a boundary point to the centroid,
slope table is used as a feature along whereas 𝑃𝑃𝑥𝑥 is the x-component and
with the other multiple features for 𝑃𝑃𝑦𝑦 is the y-component of the line P.
face recognition. A slope table is 𝑃𝑃𝑥𝑥 and 𝑃𝑃𝑦𝑦 are used to compute the
constructed by calculating the
slope as shown below.
slopes of the different fiducial
𝑃𝑃𝑦𝑦
landmarks of facial components 𝑚𝑚 = tan 𝜃𝜃 =
𝑃𝑃𝑥𝑥
(20)
(left eye, right eye, nose, and lips).
Table II and Table III show the these approaches use geometrical
results for features computed using features and the motion vector-
all the above elaborated approaches based interpolation technique. The
applied on different numbers of performance of RASF is best
persons. To test these approaches, because of the addition of a slope
the data set was divided into sub- feature in the features of the RAF
parts. Hence, all the approaches approach. So, it can be concluded
were tested on 2 persons, then 3, 4, that the features of the slope table
and 5 persons, which helped to are working as very useful face
understand the dimensionality and recognition features, achieving the
generalization of the current highest accuracy in all the sub-parts
approaches. of the data set.
Table II depicts the performance The average percentage
in terms of percentage accuracy of accuracy of classifiers based on all
Approach 5 (RASF) using NNge as approaches shows that the Support
an effective approach than the rest Vector Machine (SVM) classifiers
of the four techniques and six outperform the rest of the classifiers
classifiers for 2 persons. However, including Neural Network
for persons 3 and 5, RASF with (Multilayer Perceptron (MLP)),
SVM classifiers performs better. Decision Table, AdaBoost, Naive
For person 4, the best approach Bayes, J48 and NNge. It is because
remains RASF but with MLP SVM has a regularization parameter
classifiers. Moreover, the average which helps in avoiding overfitting
percentage accuracy of Approach 4 and uses the kernel trick, which
(RAF) using all classifiers for builds expert knowledge about the
persons 2 and 5 is remarkably better problem via engineering the kernel.
than the rest of the approaches, The second and third-best average
while the average percentage percentage accuracy are shown by a
accuracy of RASF for persons 3 and neural network (Multilayer
4 is best among all approaches. Perceptron (MLP)) and Naive
Hence, it can be summarized that Bayes classifiers, respectively (see
RAF and RASF show a clear Table II). The grounds of improved
generalized behavior that percentage accuracy of these
outperforms all the other classifiers must be the ability to
approaches, irrespective of the handle the missing values which is
number of persons. The possible known as completeness property.
reasons for the better performance This property assures the
of RAF and RASF are that both of scrutinizing of all possible
School of System and Technology
101
Volume 2 Issue 2, Fall 2022
Feature Based Techniques for...
Table 3 shows the results for the efficient classifier. On the other
time taken in building the model for hand, neural network (MLP)
the features computed and tested consumes the most time out of all
using the above mentioned the classifiers, whereas SVM
approaches. It is clear from the overtakes all the other classifiers by
results that Naive Bayes is the most achieving the highest accuracy and
time-efficient approach and does not compromise time
AdaBoost is the second most time- efficiency either.
Table III
Time taken to build a model for Random Projection (RPF), Histogram of
Oriented Gradients (HOGF), Regional Properties (RePF), Ratio and
Angles (RAF), and Ratio, Angles, and Slope (RASF) with Different
Classifies
Person SVM AdaBoost NNge N-B D-Table J48 MLP
1 10 30 10 0 40 10 3680
2 30 10 30 0 40 40 5330
RPF
3 40 10 30 0 60 50 7390
4 80 10 60 10 80 80 10050
1 10 10 10 0 10 0 90
2 20 0 0 0 10 0 150
HOGF
3 40 0 10 0 20 10 620
4 50 0 0 0 20 0 350
1 10 10 10 10 40 10 2150
2 20 30 10 0 60 10 3080
RePF
3 30 10 30 0 60 20 4090
4 60 10 40 0 110 20 5830
1 10 30 10 0 60 10 4470
2 20 50 20 0 120 10 6440
RAF
3 40 10 40 0 160 30 8760
4 70 10 60 0 270 50 9940
1 60 130 220 40 330 80 2190
2 30 40 20 0 80 20 3030
RASF
3 40 10 30 0 100 20 4110
4 70 10 30 0 140 80 5540
Average 37 21 33.5 3 90.5 27.5 4364.5
Naïve Bayes=N-B, Decision table=D-Table
Fig. 11(a-e) shows the pictorial Projection (RPF), Histogram of
representations of the results of the Oriented Gradients (HOGF),
approaches used namely Random Regional Properties (RePF), and
Fig. 10. Graph showing the results of Random Projection (RPF), Histogram
of Oriented Gradients (HOGF), Regional Properties (RePF), Ratio and
Angles (RAF), Ratio, Angles, and Slope (RASF) in terms of percentage
accuracy with 10-fold cross-validation
in Proc. Second Doct. Symp. Sci., vol. 11, no. 24, Art. no.
Comput. Intell., vol. 40, no. 28, 11600, Dec. 2021, doi:
2021, pp. 654–668, doi: https://doi.org/10.3390/app1124
https://doi.org/10.1007/978- 11600
981-16-3346-1_9 [36] S. F. Ali and M. T. Hassan,
[32] X. Zhu, X. Liu, Z. Lei, and “Feature based techniques for a
S. Z. Li, “Face alignment in full driver’s distraction detection
pose range: A 3d total solution,” using supervised learning
IEEE Trans. Pattern Anal. algorithms based on fixed
Mach. Intell.,vol. 41, no. 29, pp. monocular video camera,” KSII
78–92, Nov. 2017, doi: Trans. Internet Info. Syst., vol.
https://doi.org/10.1109/TPAMI. 12, no. 8, pp. 3820–3841, Aug.
2017.2778152 2018.
[33] A. Mollahosseini, B. Hasani, [37] S. F. Ali, M. Muaz, A.
M. J. Salvador, H. Abdollahi, D. Fatima, F. Idrees, and N. Nazar,
Chan, and M. H. Mahoor, “Human fall detection,” in
“Facial expression recognition Inmic. IEEE, 2013, pp. 101–
from world wild web,” in Proc. 105.
IEEE Conf. Comput. Vision [38] H. Du, H. Shi, D. Zeng, X.-
Pattern Recog. Workshops, vol. P. Zhang, and T. Mei, “The
42, no. 29, 2016, pp. 78–92. elements of end-to end deep face
[34] M. Valstar, B. Martinez, recognition: A survey of recent
X. Binefa, and M. Pantic, advances,” ACM Comput.
“Facial point detection using Surveys (CSUR), vol. 54, no.
boosted regression and graph 10s, pp. 1–42, 2022, doi:
models,” in Comput. Vision https://doi.org/10.1145/350790
Pattern Recog., 2010 IEEE 2
Conf. IEEE, June 13–18, 2010, [39] G. Jeevan, G. C. Zacharias,
pp. 2729–2736, doi: M. S. Nair, and J. Rajan, “An
https://doi.org/10.1109/CVPR.2 empirical study of the impact of
010.5539996 masks on face recognition,”
[35] S. F. Ali, A. S. Aslam, M. J. Pattern Recognition, vol. 122,
Awan, A. Yasin, and R. Art. no. 108308, Feb. 2022, doi:
Damaˇseviˇcius, “Pose https://doi.org/10.1016/j.patcog.
estimation of driver’s head 2021.108308
panning based on interpolation [40] M. Smith and S. Miller,
and motion vectors under a “The ethical application of
boosting framework,” Appl. biometric facial recognition