Robust Face Detection and Recognition Using

Image Processing and OpenCV

Aruna Bhat Rahul Kumar Jha Vansh Kedia
Department of Computer Engineering Department of Computer Engineering Department of Computer Engineering
Delhi Technological University Delhi Technological University Delhi Technological University
Delhi, India Delhi, India Delhi, India
aruna.bhat@dtu.ac.in heavenboyrj.11@gmail.com vanshkedia1999@gmail.com

Abstract— We have witnessed enormous progress in the the aid of Python and OpenCV, the technology is the most
fields of Artificial Intelligence, Machine Learning, and efficient method of detecting the face of a person. This
Computer Vision over the last decade, owing to a doubling in technology has applications across a wide range of domains,
processing capacity per year, as well as a massive rise in data including the military, security, schools, colleges, and
internationally. Facial identity, biometrics, and face universities, airplanes, banking, online web applications,
identification have gone from a rarefied issue to a prominent gaming, and other similar areas” [10]. Facial detection is not
topic for academics in computer science these days. Due to the
the same to face recognition. Facial detection is the process
basic nature of the topic, breakthroughs in computer image
processing and understanding studies will give insights into
of recognizing and finding a person's face in a photograph.
how the human brain operates, according to the common The method of recognizing a person's face from a database
belief. Face detection and recognition are often employed in of photos is known as face recognition. Thanks to the
national defense, CCTV surveillance, criminal investigation, media, the use of photographs to identify persons has
personal protection, law enforcement, biometrics, and a variety become commonplace. It is, however, more sensitive to
of other applications. For our project, we utilized OpenCV, fingerprint and retina scanning than other types of
Haar cascade, NumPy, and Matplotlib Python packages. biometrics[9]. “Among the features of the face recognition
Haar-Cascades will be utilized for face detection, and OpenCV system are faster processing, identification automation,
provides three built-in face recognizers: Face recognition invasion of privacy, large data storage, best results,
algorithms such as Eigenfaces, Fisher faces, and Local binary increased security, real-time face recognition of students and
pattern histograms are used. The approach employed in this employees in schools and colleges, employees in corporate
research is literature review, which entails reading and offices, smartphone unlock, and a host of others” [8]. Face
examining numerous books or literature relevant to detection is one of the most complex problems in the real
mathematical ideas that underpin the development of
world scenario because the same person is detected as a
algorithms to detect a human's face, which are subsequently
followed by implementation in code, particularly Python code.
different person or detection fails depending upon
expression, light, pose. Being a part of biometric
Keywords— Face Recognition, Deep neural Network, technology, face detection has become one of the most
histograms, matplotlib, OpenCV, Image Processing important topics in the machine learning field. Facial
recognition software is used in a variety of areas, including
I. INTRODUCTION entertainment, security, law enforcement, and biometrics. It
Face detection is a kind of CV technology that assists in the has evolved over time from a clumsy computer vision
visualization of individual’s faces and is concerned with the approach to a powerful artificial neural network. Face
recognition of instances of objects of a certain type in digital recognition is essential for face analysis, tracking, and
photos and real time images, among other things[11]. It's all identification[7].
about detecting faces in an image or video frame and II. PROBLEM STATEMENT
extracting them so that face recognition algorithms may
utilize them. Face recognition has become more important Through this research project, we plan on gaining a better
as processing power, data storage, and technology have understanding of OpenCV and Face Detection and Face
improved, particularly in industries such as security, CCTV Recognition using OpenCV libraries. We also intend to
surveillance, biometrics, social media, and other comparable present a comparative study of various face detection and
disciplines. Face recognition is a method for determining a recognition frameworks. The objective of our project is to
person's identification based on their unique facial detect human faces from input colored image. Here are few
features[12]. It's all about extracting characteristics that are objectives to design face detection system [15]. The
uniquely described in the face from a Grayscale picture that objective of face detection is:
has been retrieved, scaled, cropped, and transformed. ● Create a real-time facial recognition system [16].
Photos, films, and real-time machines may all benefit from ● To use a Haar Classifier-based facial recognition
such systems[13]. Face recognition is a non-invasive
method.
identification technology that can analyse numerous faces at
the same time, making it quicker than previous methods. ● To use OpenCV to create a face detection system.
“An individual's face may be detected with relative ease A fascinating and challenging problem in the field of AI is
using such technology, which makes use of an existing CV, which is both interesting and difficult to solve.
dataset that has similar look to that of the individual. With Computer Vision links software to the world around us. It

XXX-X-XXXX-XXXX-X/XX/$XX.00 ©20XX IEEE

lets software comprehend and learn from its surroundings. objects, it was primarily driven to detect faces. This
Face detection and identification has risen from obscurity to framework can analyze pictures quickly and achieve
become one of the most prominent fields of computer vision excellent detection rates” [9]. Face Detection system based
research in the recent decade [14]. Due to the basic nature of on retinal connected neural network (RCNN): Different
the topic, breakthroughs in computer image processing and ANN architectures and models have been employed for face
understanding studies will give insights into how the human identification in recent years. “According to Rowley, Baluja,
brain operates, according to the common belief. It is gaining and Kanade, a retinal connected neural network (RCNN)
a lot of traction throughout the globe due to its contactless analyzes small windows of an image to evaluate whether
biometric capabilities. Traditional fingerprint scanners are each window contains a face “[12].
being phased out in favor of AI-based technology, which
has enormous business implications. Its applications include
security and surveillance, authentication, access control,
digital healthcare, image retrieval, and more. As we have IV. METHODOLOGY
seen in recent times Face Detection has gained a lot of We used the OpenCV package and approach to create these
attention from researchers due to its application in computer in Python. Haar Cascades will be utilized for face detector,
and human communication. Face detection is a part of while Eigenfaces, Fisher faces, LBP and LBPH will be
image processing [2-4]. Face detection is a technique for employed for face recognizer. Firstly, we will create a face
compressing, improving, or extracting features images and detection using Haar-cascades then for face recognition we
identifying single or many faces from pictures, removing will be implementing three algorithms Eigenface, Fisher
unwanted background noise. For this, the algorithm face and Linear binary pattern histograms. Additionally, we
basically categorizes images into two groups based on the will be using python deep learning algorithms and more
presence of a face in it or not. The algorithm will be depending on faster, accurate and efficient results [5-6].
thoroughly going through the images and then identify the
A. Tools and Technologies
presence of the face. Face Recognition technology is one of
the most relevant biometric technologies. The technique 1) HAAR Cascode
gets benefits from the most visible human body components Viola-Jones Face detection, also known as Haar Cascades, is
in a non-interrupting way. According to worldwide data, a ML approach for training a cascade function using many
most individuals are unaware of the face recognition process positive instances (such as photographs of people's faces)
that is taking place on them, making it one of the least and negative instances (like images that are without face)
invasive procedures with the least amount of delay. The face [1-3]. The Haar Cascade is a common face detection
recognition algorithm goes through the facial features of the algorithm for photos and videos. Image characteristics are
input image and training image features for the recognition. handled as numeric data collected from the training and
This biometric has been considered, and perhaps testing photos and utilized to discriminate between them to
exaggeratedly, lauded as a great system for relating possible recognize faces.
troubles like terrorists, fiddle artists, and so on, but it has yet
to gain wide acceptance in high-position use. Biometric face
recognition technology is anticipated to surpass point
biometrics as the most common system of stoner
identification and authentication soon.

III. LITERATURE REVIEW

This section gives an overview of the most common human
face recognition methods, which are usually used on frontal
faces, as well as the advantages and disadvantages of each
approach. The approaches examined include eigenfaces,
CNN, Deep Learning, dynamic link architecture, hidden
Markov model, geometrical feature matching, and template Fig.1 Haar cascade flowchart
matching. The techniques are compared based on the face
representations they use. Skin color-based Face detection We have applied every feature of the algorithm on all
Method: “Face detection in color photographs is a training images by giving equal weights at the starting. It
widespread and successful method. Skin color is vital. determines the valid threshold for positive and negative
Using skin color to trace a face has several advantages. pictures. There are chances of error and misprediction, so
Color processing requires far less time than other facial we have selected features with a minimum rate of error and
processing. Each pixel was classified as skin or non-skin better algorithm performance. They are the features that are
using the skin color detection technique” [11]. Viola and best for image classification.
Jones Face Detection system “Paul Viola and Michael Jones
introduced the Viola–Jones object identification framework
in 2001. Although it may be taught to recognize a range of
 Fig.2 Haar kernel

The image's sample representation is shown on the left, with

Fig.3 Flowchart of the Eigenfaces Method
pixel values ranging from 0 to 1. The picture in the middle
displays a haar kernel with all bright pixels set to 0 and all
dark pixels set to 1.” An image's haar is computed by Eigenfaces method is widely used as it is independent upon
comparing the average of pixels in darker and brighter facial geometry, simplicity of realization, fast recognition
areas, the sum of all picture pixels in the darker half of haar compared to other approaches, high success rate etc.
feature, as well as all image pixels in the lighter half of haar
feature” [2]. The next step is to look for differences, and if 3) FisherFaces
the difference is near to one, the edge of the difference is Being one of the most popular and superior algorithms,
found. The picture characteristics are calculated using all of Fisher Faces has been in working on facial recognition for a
each kernel's available sizes and pixel positions. There are long time. Fisher Face algorithm method for face
two primary states of the Haar cascade image classifier one recognition is a merger of PCA and LDA [20-21]. The
is training and another one is the detection, and both are Principal Component Analysis (PCA) is used for the
provided by OpenCV [17]. reduction of face space, size, pixel, and the dimension of the
The structure of the face recognition system is end to end image. Dimension reduction during PCA reduces
recognized. The different convolutional layers are used in discriminant information relevant for LDA. Then, Fisher’s
the system of face recognition and detection. The layer with Linear Discriminant (FDL), commonly known as Linear
the size is tabulated as shown in table.1 Discriminant Analysis (LDA), is used to extract picture
properties. Face picture matching using minimal Euclidean
[18].
layer Size in Size out kernel
4) Local Binary Patterns Histogram(LBPH)
L2 1 1 128 1 1 128 Local Binary Patterns Histograms is one of the powerful
Lc3 1×32×128 1×1×128 algorithms for texture classification. LBPH detects pixels in
Lc2 1×32×128 1×32×128 maxout p=2 a picture by thresholding their neighbors. In other words,
LBPH compares each pixel to its neighbors to summarize an
Lc1 512×7×7 1×32×128 maxout p=2
image's local structure and converts the result to a binary
Table.1 structure of network layers involved
integer [6].
2) EigenFaces
Principal Component Analysis underpins the Eigenfaces
approach (PCA). For the efficient representation of facial V. RESULT ANALYSIS
pictures, Sirovich and Kirby employed Eigenfaces and PCA.
Fisher Faces algorithm considers faces from all the training
Turk and Pentland went on to apply it for facial recognition
images at once and finds principal components from all of
difficulties. PCA is frequently used in pattern recognition
them at once. Focusing on characteristics that represent all
because it is a technique of subspace projection with low
the faces in the training data rather than distinguishing one
noise sensitivity. The Eigenfaces method's fundamental
from another. As it finds illumination as a relevant feature of
approach is to extract the face's distinguishing
image and takes variation in illumination as principal
characteristics and represents linear combination of
component of the input images and discards other relevant
"eigenfaces" that are acquired during extracting features
features that are useful for recognition, we have used
phase, followed by the computation of facial elements in the
Eigenfaces to extract such features and the issue is fixed.
training set. Because not all elements of the face are equally
Fisher Faces just prevents features of one's image/face
significant for facial identification, Eigenfaces algorithms
becoming dominant on other images.
take that into account. The major component evaluates all of
the training photos and extracts key and useful information
from them. Following that, the input picture characteristics
are compared to the main components (relevant and useful
features taken from training photos) that were saved
throughout the training phase [19]. The face is recognized
by projecting it into the space generated by the eigenfaces.
Then comparing the Euclidean distance between the
eigenface’s eigenvectors and the eigenfaces to the picture.
The recognition is completed if the distance is near to zero.
 Fig.4 Face Image with variance in illumination the brightness and contrast of your face photos. “To improve
our accuracy, we could use HSV or other color space instead
One of the major problems of using Fisher Faces method is of RGB, or add image processing processes like edge
computation problems as it has a sophisticated computing enhancement, contour detection, motion detection, etc. [2].
process [22-23]. Also, the condition of face images is We're also scaling photos to a standard size, which may alter
affected due to the variation of light (as shown in above the face's aspect ratio. The picture may be resized via
figure the algorithm will extract just one individual's Facial interpolation without losing its aspect ratio. This approach
features) and different expression and position of face image makes advantage of a neighborhood of pixels to enlarge and
(as shown in figure below the algorithm has extracted reduce the image's size. For face detection, OpenCV’s Haar
different features of one person only). Cascade classifier was employed [25]. Based on the input
image, which might be from a file, a camera, or live video,
The facial identification algorithm evaluates each picture
Fig.5 Grayscale Images
position and labels it as "Face" or "Not Face". Picture
resolution of 100 x 100 pixels is used to classify the face.
The combination of LBP and HOG descriptors improves Faces in the picture may be smaller or bigger; the classifier
face detection performance [4]. LBPH aims to extract local scans the picture many times looking for individual’s faces
characteristics from photos. Its main idea is to focus on local of varied sizes. It may seem to be a complex and
features rather than to look at the whole image as h high time-consuming operation, but thanks to computational
dimensional vector. Face image is represented as a data methods, the categorization is completed quickly, especially
vector. when done at many scales. To categorize picture locations,
the classifier consults data saved in an XML file. Before you
start writing your code, find the XML file that will be
Fig.6 LBPH algorithm working
utilized and double-check that it is accurate [26]
.
In the above figure, a center pixel value (intensity at a
certain point) of the image is taken as the threshold. A
Fig.8 training loss and loss history
pixel's intensity is expressed with a 1 if it is more than or
The fig.8 shows the training and loss history of the system
equal to its neighbor's, and a 0 otherwise. Taking these o
and 1 in an anticlockwise direction forms a binary number
and then it is converted to decimal and a new pixel value VI. CONCLUSION AND FUTURE WORK
for neighbor is defined. Circular LBP may be used to
change the radius and neighbors using bilinear interpolation To upgrade the recognition performance there are many
[8]. Then the image is divided into multiple grids to extract improvements that can be made such as the addition of
the histogram of each grid. Due to grayscale, each color processing, edge detection, increasing no of the input
histogram has 256 positions. image, etc. [24]. To increase the face recognition accuracy
more input image i.e., minimum 40 images of everyone’s
Fig.7 local Binary Pattern Histogram face, all from various viewing angles and having variation
in lighting exposures. If taking photos is not possible then
The final histogram will be representing features of the there are a variety of approaches for obtaining additional
original image. Two histograms are then compared and the input photographs, including producing new images from
image with the closest histogram is recognized. current ones: -
When it comes to picture quality, there are several variables ● By creating minor copies of facial images, we can
that impact the system's accuracy. It's critical to standardize have more training images and improved accuracy
the training photos that we feed into the face recognition and decrease the chances to biased recognition.
system by using various image pre-processing methods.
● We may resize and rotate current face photos to
Face recognition algorithms are often light sensitive, thus if
provide more alternative training data, reducing
we have taught the algorithm to detect a person in low light,
it will be difficult for it to recognize them in excellent light. sensitivity to certain situations.
This difficulty is called as "Illumination dependent," and it ● To have additional photos, we may have
includes issues like facial alignment, consistency among incorporated image noise and to increase noise
pictures, consistency in size, rotation angle, hair and tolerance.
expression, and lighting. That is why, before training your
Having a wide range of conditions is essential in an
system with training photos and implementing face
individual's training images which will help in face
recognition, it is critical to utilize excellent image
recognition in various positions and lighting conditions,
pretreatment filters. Face pixels with an elliptical mask to
else the classifier will be only looking for certain specific
expose the inner facial region, rather than the hair and
conditions else will be unable to recognize them. Also, it's
background, should be deleted. The article uses a grayscale
very important to ensure the training image should not be
image-based face identification system. The project also
too varied i.e., rotations like 90 degrees are not preferred,
explains how to convert a color facial picture to grayscale
which could make the classifier too generic and resulting in
and utilize Histogram Equalization to automatically equalize
poor accuracy so rotation more than 20 degree will end up D. Jayant and A. Bhat, "Study of robust facial
creating a separate set of training image for each person recognition under occlusion using different
with left, forward, and right images. So, for better results techniques," 2021 5th International Conference on
good preprocessing on the training image is required. Intelligent Computing and Control Systems
Histogram Equalization, as previously stated, is a very (ICICCS), 2021, pp. 1230-1235, doi:
simple picture preprocessing approach so we'll need to mix 10.1109/ICICCS51141.2021.9432300.
a few different approaches till we obtain good results. The
[2] Bhat, Aruna. "Possibility Fuzzy C-Means Clustering
algorithm's main goal is to detect faces by subtracting the
For Expression Invariant Face Recognition."
testing image from the training image and looking for
International Journal on Cybernetics & Informatics
similarities. When a person does it, it works well, but when
(UCI) 3, no. 2 (2014): 35-45.
a machine does it, it is just concerned with pixels and
numbers. So, try deleting a pixel's gray scale value in the [3] A. Bhat, S. Rustagi, S. R. Purwaha and S. Singhal,
test picture from the value of a pixel at the same coordinate "Deep-learning based group-photo Attendance
in each training image, and the smaller the difference, the System using One Shot Learning," 2020
better the match. If the background is different, since the International Conference on Electronics and
algorithm does not know what background is and what is Sustainable Communication Systems (ICESC),
foreground (face), it is crucial to get rid of as much of the 2020, pp. 546-551, doi:
background as possible, by using small sections within the 10.1109/ICESC48915.2020.9155755.
individual’s facial area that do not add any kind of
[4] X. Sun, P. Wu, and S. C. H. Hoi, “Face detection
background information. The difficult elements, such as the
using deep learning: An improved faster RCNN
backdrop, eyeglasses, beard, and mustache, are addressed.
approach,” Neurocomputing, vol. 299, pp. 42–50,
Because pictures must be correctly aligned, little
2018, doi: 10.1016/j.neucom.2018.03.030.
low-quality images often provide better recognition than
larger ones with greater resolution. Even if the training [5] S. S. Farfade, M. Saberian, and L.-J. Li,
picture is perfectly matched, if your input picture a shade “Multi-view Face Detection Using Deep
darker in compared to the training picture, the algorithm Convolutional Neural Networks Categories and
may conclude that the two images aren't a good match. In Subject Descriptors,” 5th ACM Int. Conf. Multimed.
certain circumstances, Histogram equalization may assist, Retr., pp. 643–650, 2015.
but it may make matters worse in others, thus brightness
[6] J. Jeon et al., “A real-time facial expression
discrepancies are a challenging and prevalent issue. Issues
recognizer using deep neural network,” ACM
such as a shade on right side of the eye in the training
IMCOM 2016 Proc. 10th Int. Conf. Ubiquitous Inf.
picture and a shade on the left side of the eye in the testing
Manag. Commun., 2016, doi:
image might also cause a mismatch. Face recognition is
10.1145/2857546.2857642.
best done in real-time because of this. If you're training
your algorithm with any individual’s picture and then [7] W. T. Chu and W. W. Li, “Manga FaceNet: Face
attempting to identify him/her in the moment, you'll be detection in manga based on deep neural network,”
using the same camera, environment, lighting condition, ICMR 2017 - Proc. 2017 ACM Int. Conf. Multimed.
expression, viewing direction, and location, all of which Retr., pp. 412–415, 2017, doi:
will result in high identification accuracy. However, when 10.1145/3078971.3079031.
the circumstances are different, such as in other
[8] W. Jiang and W. Wang, “Face detection and
orientations, rooms, or outdoors, or at different times of the
recognition for home service robots with end-to-end
day, it often leads to poor results. Consequently, it is critical
deep neural networks,” ICASSP, IEEE Int. Conf.
to do extensive testing to have a deeper knowledge of
Acoust. Speech Signal Process. - Proc., pp.
algorithms and obtain the best results possible. The
2232–2236, 2017, doi:
emphasis of future research will be on improving the
10.1109/ICASSP.2017.7952553.
success rate of extremely big datasets.
[9] S. Jafri, S. Chawan, and A. Khan, “Face
ACKNOWLEDGMENT Recognition using Deep Neural Network with
The Authors are indebted to central library of Delhi LivenessNet,” Proc. 5th Int. Conf. Inven. Comput.
Technological University for providing all the research Technol. ICICT 2020, pp. 145–148, 2020, doi:
materials and lab access for the proper competition of the 10.1109/ICICT48043.2020.9112543.
research work in the university. We are thankful to Our
supervisor Aruna Bhatt for guiding us in each step in the [10] J. Li and E. Y. Lam, “Facial expression recognition
research work. using deep neural networks,” IST 2015 - 2015 IEEE
Int. Conf. Imaging Syst. Tech. Proc., 2015, doi:
REFERENCES 10.1109/IST.2015.7294547.
[1] B. Ahn, J. Park, and I. S. K. B, “Real-Time Head [11] B. Yang, J. Cao, R. Ni, and Y. Zhang, “Facial
Orientation from a Monocular,” Springer, vol. 1, pp. Expression Recognition Using Weighted Mixture
82–96, 2015, [Online]. Available: Deep Neural Network Based on Double-Channel
https://link.springer.com/chapter/10.1007/978-3-319
 Facial Images,” IEEE Access, vol. 6, pp. [21] X. Li, S. Lai, and X. Qian, “DBCFace: Towards
4630–4640, 2017, doi: Pure Convolutional Neural Network Face
10.1109/ACCESS.2017.2784096. Detection,” IEEE Trans. Circuits Syst. Video
Technol., vol. 8215, no. c, pp. 1–13, 2021, doi:
[12] A. Mollahosseini, D. Chan, and M. H. Mahoor,
10.1109/TCSVT.2021.3082635.
“Going deeper in facial expression recognition
using deep neural networks,” 2016 IEEE Winter [22] B. K. Shah, A. Kumar, and A. Kumar, “Dog breed
Conf. Appl. Comput. Vision, WACV 2016, 2016, doi: classifier for facial recognition using convolutional
10.1109/WACV.2016.7477450. neural networks,” Proc. 3rd Int. Conf. Intell.
Sustain. Syst. ICISS 2020, pp. 508–513, 2020, doi:
[13] J. Kim, N. Shin, S. Y. Jo, and S. H. Kim, “Method
10.1109/ICISS49785.2020.9315871.
of intrusion detection using deep neural network,”
2017 IEEE Int. Conf. Big Data Smart Comput. [23] H. Li, S. Alghowinem, S. Caldwell, and T. Gedeon,
BigComp 2017, pp. 313–316, 2017, doi: “Interpretation of Occluded Face Detection Using
10.1109/BIGCOMP.2017.7881684. Convolutional Neural Network,” INES 2019 - IEEE
23rd Int. Conf. Intell. Eng. Syst. Proc., pp. 165–170,
[14] M. Shah and R. Kapdi, “Object detection using deep
2019, doi: 10.1109/INES46365.2019.9109524.
neural networks,” Proc. 2017 Int. Conf. Intell.
Comput. Control Syst. ICICCS 2017, vol. [24] S. Zhang, C. Chi, Z. Lei, and S. Z. Li, “RefineFace:
2018-January, pp. 787–790, 2017, doi: Refinement Neural Network for High Performance
10.1109/ICCONS.2017.8250570. Face Detection,” IEEE Trans. Pattern Anal. Mach.
Intell., vol. 43, no. 11, pp. 4008–4020, 2021, doi:
[15] H. Kim, Y. Lee, B. Yim, E. Park, and H. Kim,
10.1109/TPAMI.2020.2997456.
“On-road object detection using deep neural
network,” 2016 IEEE Int. Conf. Consum. Electron. [25] N. Do, I. Na, and S. Kim, “Forensics Face Detection
ICCE-Asia 2016, pp. 31–34, 2017, doi: From GANs Using Convolutional Neural Network,”
10.1109/ICCE-Asia.2016.7804765. no. August, 2018.
[16] B. K. Shah, V. Kedia, and R. K. Jha, “Integrated [26] A. Bhatia, V. Kedia, A. Shroff, M. Kumar, B. K.
Vendor-Managed Time Efficient Application to Shah, and Aryan, “Fake currency detection with
Production of Inventory Systems,” Proc. 6th Int. machine learning algorithm and image processing,”
Conf. Inven. Comput. Technol. ICICT 2021, pp. Proc. - 5th Int. Conf. Intell. Comput. Control Syst.
275–280, 2021, doi: ICICCS 2021, no. Iciccs, pp. 755–760, 2021, doi:
10.1109/ICICT50816.2021.9358504. 10.1109/ICICCS51141.2021.9432274.
[17] Bhat, Aruna. "Makeup invariant face recognition [27] B. K. Shah, A. K. Jaiswal, A. Shroff, A. K. Dixit, O.
using features from accelerated segment test and N. Kushwaha, and N. K. Shah, “Sentiments
eigen vectors." International Journal of Image and Detection for Amazon Product Review,” 2021 Int.
Graphics 17, no. 01 (2017): 1750005. Conf. Comput. Commun. Informatics, ICCCI 2021,
2021, doi: 10.1109/ICCCI50826.2021.9402414.
[18] Bhat, Aruna. "Robust Face Recognition by
Applying Partitioning Around Medoids Over Eigen [24] M. Kumar, A. Bhatia, P. Gupta, K. Jha, R. K. Jha,
Faces and Fisher Faces." International Journal on and B. K. Shah, “S IGN L ANGUAGE A
Computtaional Sciences & Applications 4, no. 3 LPHABET R ECOGNITION U SING,” no. Iciccs,
(2014). pp. 1859–1865, 2021.
[19] Bhat, Aruna. "Medoid based Model for Face [25] V. Kedia, A. Bhatia, S. R. Regmi, S. Dugar, K. Jha,
Recognition using Eigen and Fisher Faces." and B. K. Shah, “Time Efficient IOS Application
Available at SSRN 3584107 (2013). For CardioVascular Disease Prediction Using
Machine Learning,” no. Iccmc, pp. 869–874, 2021.
[20] A. Shroff, B. K. Shah, A. Jha, A. K. Jaiswal, P.
Sapra, and M. Kumar, “Multiplex Regulation [30] B. Ahn, J. Park, and I. S. K. B, “Real-Time Head
System with Personalised Recommendation Using Orientation from a Monocular,” Springer, vol. 1, pp.
ML,” Proc. 5th Int. Conf. Trends Electron. 82–96, 2015, [Online]. Available:
Informatics, ICOEI 2021, pp. 1567–1573, 2021, doi: https://link.springer.com/chapter/10.1007/978-3-319
10.1109/ICOEI51242.2021.9453005. -16811-1_6.