International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056

Volume: 09 Issue: 07 | July 2022 www.irjet.net p-ISSN: 2395-0072

Age and Gender Classification using Convolutional Neural Network

Ms. Suhasini1, Ms. T C Apoorva2, Mr. Vinay M3, Mr. Sunil M S4, Ms. Harshitha M N5

1 Assistant Professor, Dept. of Computer Science and Engineering, Maharaja Institute of Technology,Thandavapura
2,3,4,5 Students, Dept, of Computer Science and Engineering, Maharaja Institute of Technology, Thandavapura
------------------------------------------------------------------***-----------------------------------------------------------------
Abstract - Due to its numerous applications in impact on facial recognition systems. This concept is
various facial analysis challenges, automatic prediction of crucial to the new fields of computer vision research that
age and gender from face images has received a lot of will be investigated.
interest recently. The available models, however, are still
below the needed accuracy level, which is required for the 1.1 Overview with Problem Statement
usage of these models in real-world applications due to
the significant intra-class variance of face images (such as With the expansion of real-world applications has
difference in lighting, position, scale, and opacity). In this expanded day-to-day living, researchers have shown
study, we offer a classification model that can accurately more interest in the soft biometrics sector to close the
identify the gender and age range of facial images using communication gaps between humans and machines.
convolutional neural networks Age, gender, ethnicity, height, face dimensions, and other
soft biometrics are included.
Key Words: Convolutional Neural Network, Machine
learning, Age classification, Gender Detection. Machines cannot classify patterns as effectively and
powerfully as the human brain can. Therefore, our goal is
1. INTRODUCTION touse technology to imitate the ability of the human brain
to determine a person's age and gender. This problem can
Age, gender, mood, and other characteristics can be solved by developing an application for age and
all be inferred from a person's face. Numerous dynamic gender detection that can accurately determine a
aspects that alter over time, such as age, hairstyles, person's age and gender. The age and gender of the person
expressions, etc., have an impact on it. Age and gender are are determined by using their human face as the input.
regarded as crucial biometric characteristics for The person's age and gender are the output.
identifying humans. For the purpose of human
1.2 Challenges and Applications
identification and verification, biometric recognition
gathers data on a person's physiological and behavioral Age estimation and facial gender classification provide
traits (security models). Age, gender, ethnicity, height, numerous difficulties. Two classes that can be either male
and face measurements are examples of soft biometrics. or female are subject to gender prediction. While a
Hard biometrics are measurements of the body (physical, machine cannot easily classify gender, a human can.
behavioral, and biological). To speed up data traversal or Numerous methods and models have been criticized for
to categorize unlabeled subjects for different gender and gender classification based on extra data from hairstyles,
age groups, soft-biometric features can be retrieved, such body shape, attire, and facial traits. As of now, it is not
as skin color, hair color, the distance between the eye and possible to determine actual age while estimating age. In
nose,facial shape, etc. order to determine age from facial photos, age grouping
is still used. Additionally, there aren't enough high-quality
Additionally, with the ubiquitous use of datasets for estimating age and classifying gender to
computers, biometric identification is becoming more and support extensive research.
more necessary in sectors like healthcare and home
automation. Through pattern recognition, computer Partial occlusions and poor-quality photos are the most
vision, and picture analysis, it has recently become frequent issues when it comes to face detection or
possible to automatically determine one's physical age/gender classification. Due to the model's limited data
presence and verify their identification. Age is one of the set and difficulty in making predictions, these directly
biometric characteristics taken into account. Numerous affect the outcome findings. When a human is making the
factors, including DNA alteration, metabolic changes, sun forecast, the same rules still apply. It is more difficult for a
exposure, variations in face tissues, reorganization of the human to comprehend what is being viewed in a low-
facial bones, and others, contribute to aging. The aging quality image and, as a result, to predict the future.
of the face has a negative

© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 277
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 07 | July 2022 www.irjet.net p-ISSN: 2395-0072

Age recognition is important in police investigations and • Pooling Layers: It reduces the spatial dimensions
intelligence departments because it aids in locating the of the input volume for the next convolutional layer. It
actual suspect based on his age. They may receive a only affects the length and height, and not the depth.
filtered result of that person who has committed a
criminal act or any other activity. When it comes to • Fully connected Layers: Several connected
software, the actual and predicted ages are roughly the neural network layers have been used to perform any
same, indicating its dependability, and this dependability high-level reasoning.
serves as a trust factor for many other useful operations
in daily life. There are some difficulties that might also arise with
CNN, such as overfitting, which is due to the CNN
We propose a scheme in this paper to bridge the gap
training of a large number of parameters. The solution
between automatic face recognition and age and gender
to it is the pretraining of parameters, which accelerates
prediction. When there is a large-scale improvement in
the learning process of the model as well as improves
face recognition, a link between face recognition and
the generalising capability of the model. In short, CNN
Convolution Neural Network (CNN) is proposed, and by
has outperformed the usual and traditional machine
studying it further, we created a system in which a
learning algorithms.
limited number of face data sets are used to accurately
predict ageand gender.
3. IMPLEMENTATION
2. SYSTEM ANALYSIS AND DESIGN Python programming language, as well as
numerous computer vision and machine learning
2.1 Machine learning techniques:
packages and libraries, will be used throughout the
research implementation. The primary goal of the
The various machine learning approaches
project will be to create a Python-based, Tensorflow-
utilized for categorization and implementation in this
system are covered in this section. The optimal model compatible high-level convolutional neural network API.
for the system has been determined by comparing the Python, in and of itself, is a high-level programming
results of all of these models. language. Python is open-source, object-oriented, and
has simple readability and coding. Because it contains so
Deep learning techniques for computer vision: many packages, it is widely used in Big Data, Machine
Learning, and Computer Vision. Furthermore, Python
In recent times, deep learning techniques was chosen for this experiment because it is free to use,
proved to be a big success in the Computer Vision compatible with the Windows operating system, and
discipline. Deep learning enables multi-layered contains all of the necessary libraries for face recognition,
computing models to determine and interpret data with emotion detection, and gender classification.
multiple abstraction levels and imitates how the
information is perceived and translated by the brain. So, Face identification consists of three steps. Detect
it implicitly captures large-scale data structures. Deep which part of an image is the face, then train our
learning has outperformed many of the previously classifier for that dataset of images, and finally, predict
existing techniques. Deep learning has enabled various the face. OpenCV, a Python open-source library for
techniques in computer vision to increase accuracy and computer vision, will be used for this. A Haar cascade
efficiency, which includes object detection, action frontal face default classifier will be used for face
recognition, human emotion recognition, and others. detection, which is a pretrained model that is freely
Further, types of deep learning techniques will be available online. A default Haar classifier will be used for
discussed with their comprehensive details. gender and age prediction. The first step will be to detect
the face in the image using some test images. All of the
Convolutional Neural Networks(CNN): images used in the training are freely available online
and open-source.The entire experiment will be
Convolutional Network Network (CNN) was implemented in Keras with Tensorflow as the backend.
first proposed in 1962, and it consists of the following The entire Convolutional Neural Network will be built on
layers. these, as well as the OpenCV computer vision library. Keras
can determine whether the model's current epoch
• Convolutional Layer: As it is known that CNN outperformed the previously saved epoch. In this case,
utilizes various kernels, so the convolutional operation the best model weights will be saved in a file that will
of the layer increased the learning time of the developed allow the weights to be loaded directly without
retraining if the model needs to be used in another
model.

© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 278
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 07 | July 2022 www.irjet.net p-ISSN: 2395-0072

situation. Keras has modularity, extensibility, and

Python nativeness when compared to other similar
libraries.

Gender identification is the process of

determining whether a person is male or female. It is a
binary model because there are only two options: male
or female. The primary libraries for Gender
Classification are OpenCV andKeras.

4. RESULTS

By analyzing human facial features in real-time, this

model can predict ages ranging from 0 to 80 and classify
genders as Male or Female. Because the model predicts Figure 1: Age and Gender prediction of age range 60-100
age in real-time, it is subject to change with each webcam
frame. Figure 1,2 and 3 shows the prediction results of the
developed model.

5. CONCLUSIONS

A thorough literature review of various Machine

Learning and Deep Learning techniques is used to
discuss all of the techniques and methods that have
already been implemented in this field. Facial images
have become increasingly important in recent decades,
owing primarily to their promising real-world
applications in a variety of emerging fields. The
proposed system is capable of classifying gender as
either male or female and predicting age from 0 to 80.

The model's accuracy is calculated separately to

provide a more accurate comparison and interpretation
of the study. The proposed architecture was built
methodically to improve accuracy and reduce the
Figure 1: Age and Gender prediction of age range 0-2
number of parameters. Gender classification and age
prediction have been manually tested, and the results
have been astounding. Because gender classification is
considered a binary problem in this study, it has proven
to be very efficient with the use of Keras and achieves an
overall accuracy of about 90%. Age prediction is affected
by a variety of external factors, including lighting effects,
facial expressions, and skin tones, but it also produces
impressive results.

ACKNOWLEDGEMENT

We would like to express our sincere gratitude

to Prof. Suhasini, who served as our project advisor.
Prof. Suhasini directed us through this project, offered
usimportant advice, and helped us develop it beyond our
expectations. Second, we would like to express our
gratitude to Dr. Ranjith K N, our project coordinator, for
Figure 1: Age and Gender prediction of age range 8-12 her ongoing support and assistance in helping us for

© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 279
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 07 | July 2022 www.irjet.net p-ISSN: 2395-0072

completing this project within the allotted time. Jadavpur University, Kolkata. 700032, India,
Additionally, we would like to extend our sincere Department of Information Technology, Jadavpur
gratitude to our department's head, Dr. Ranjith K N, for University, Kolkata 700106, India , Institute of
giving us access to a platform where we can attempt to Industry Revolution 4.0, The National University of
work on project development and illustrate the real- Malaysia (UKM), Selangor 43600, Malaysia, Institute
world applications of our academic curriculum. We for Mathematical Research, University Putra
would like to thank Dr. Y T Krishne Gowda, our principal, Malaysia, Serdang 43400, Malaysia
for giving us the chance to complete this wonderful
project on the topic of "Age and Gender Classification [7] Insha Rafique, Awais Hamid, Sheraz Naseer,
Using Convolutional Neural Network." This project has Muhammad Asad, Muhammad Awais, Talha Yasir, Age
also assisted us in conducting extensive research and and Gender Prediction using Deep Convolutional
learning how to implement it. Neural Networks, Department of Software
Engineering University of Management and
REFERENCES Technology Lahore, Pakistan.

[1] Asad Mustafa, Kevin Meehan, Gender Classification [8] Md. Nahidul Islam Opu, Tanha Kabir Koly, Annesha Das
and Age Prediction using CNN and ResNet in Real- and Ashim Dey, A Lightweight Deep Convolutional
Time, Department of Computing Letterkenny Neural Network Model for Real-Time Age and Gender
Institute of Technology Letterkenny, Ireland, ORCiD: Prediction, Department of Computer Science &
0000-0001- 5447-5878. Engineering Chittagong University of Engineering and
Technology Chittagong-4349, Bangladesh
[2] Thakshila R, Kalansuriya, Anuja T, Facial Image
Classification Based on Age and Gender, Dharmaratne [9] Azliza Mohd Ali, Plamen Angelov, Gender and Age
University of Colombo School of Computing, Classification of Human Faces for Automatic Detection
University of Colombo No 35, Reid Avenue, Colombo7. of Anomalous Human Behaviour, 2017 3rd IEEE
International Conference on Cybernetics (CYBCONF),
[3] Vijay Prakash Dwivedi, Deepak Kumar Singh, Saurav DOI: 10.1109/CYBConf.2017.7985780
Jha, Ranvijay, Gender Classification of Blog Authors:
With Feature Engineering and Deep Learning using [10] Seok Hee Lee, Hyuk Jin Kwon, Hyung Il Koo, Nam Ik
LSTM Networks, Computer Science and Engineering Cho, Age and gender classification using wide
Department MNNIT Allahabad UP, India. convolutional neural network and Gabor filter, 2018
International Workshop on Advanced Image
[4] Mohammed Kamel Benkaddour, Sara Lahlali, Maroua Technology (IWAIT), DOI:
Trabelsi, Human Age and Gender Classification using 10.1109/IWAIT.2018.8369721.
Convolutional Neural Network, University Kasdi
Marbah, Department of Computer Science and [11] Xuan Liu, Junbao Li, Cong Hu, Jeng-Shyang Pan, Deep
Information Technology, FNTIC Faculty, Ouargla, convolutional neural networks-based age and gender
Algeria. classification with facial images, 2017 First
International Conference on
[5] Hiromi Kondo, Fumiyo N. Kondo, Convolutional
Electronics Instrumentation & Information Systems
Neural Networks on Multichannel Time Series of
(EIIS), DOI: 10.1109/EIIS.2017.8298719.
Smartphone Applications for Gender or Age Range
Classification, Toyota Motor Corporation Customer
[12] Mohammed Kamel Benkaddour, Sara Lahlali, Maroua
First Promotion Group C&A Operations Div. Aichi,
Trabelsi, Human Age and Gender Classification using
Japan, University of Tsukuba Division of Policy and
Convolutional Neural Network, 2020 2nd
Planning Sciences Faculty of Engineering, Information
International Workshop on Human-Centric Smart
and Systems Ibaraki, Japan.
Environments for Health and Well-being (IHSH).
[6] Avishek Garain , (Member, Ieee), Biswarup Ray ,
[13] Gil Levi, Tal Hassncer, Age and gender classification
Pawan Kumar Singh , (Member, Ieee), Ali Ahmadian ,
using convolutional neural networks, 2015 IEEE
(Member, Ieee), Norazak Senu , And Ram Sarkar ,
Conference on Computer Vision and Pattern
(Senior Member, IEEE), A Deep Learning Model for
Recognition Workshops
Classification of Age and Gender From Facial Images,
DIO:10.1109/CVPRW.2015.7301352.
Department of Computer Science and Engineering,

© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 280
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 07 | July 2022 www.irjet.net p-ISSN: 2395-0072

[14] AvishekGarain, Biswarup Ray, Pawan Kumar Singh, BIOGRAPHIES

Ali Ahmadian, Norazak Senu, Ram Sarkar, A Deep
Learning Model for Classification of Age and Gender Suhasini, Professor at Maharaja
From Facial Images, DOI: Institute Of Technology
10.1109/ACCESS.2021.3085971. Thandavapura, Mysore,
Department of Computer Science
[15] Syed Taskeen Rahman, Asiful Arefeen, Shashoto and Engineering.
Sharif Mridul, Asir Intisar Khan, Samia Subrina,
Human Age and Gender Estimation using Facial
Image Processing, 2020 IEEE Region 10 Symposium
(TENSYMP), DOI: Sunil M S, Student of Maharaja
10.1109/TENSYMP50017.2020.9230933. Institute Of Technology
Thandavapura, Mysore. Pursuing
[16] Sandeep Kumar, Sukhwinder Singh, Jagdish Kumar, A Bachelor's of Engineering Degree
in Computer Science and
study on face recognition techniques with age and
Engineering.
gender classification, Published in: 2017
International Conference on Computing,
Communication and Automation DOI:
10.1109/CCAA.2017.8229960. Vinay M, Student of Maharaja
Institute Of Technology
[17] SandeepKumar, Sukhwinder Singh, Jagdish Kumar, A Thandavapura, Mysore. Pursuing
study on face recognition techniques with age and Bachelor's of Engineering Degree
gender classification, 2017 International Conference in Computer Science and
on Computing, Communication and Automation. DOI: Engineering.
10.1109/CCAA.2017.8229960.

[18] Jun Beom Ko, Wonjune Lee, Sung Eun Choi, Jahie Kim,
A gender classification method using age information, T C Apoorva, Student of Maharaja
Institute Of Technology
2014 International Conference on Electronics,
Thandavapura, Mysore. Pursuing
Information and Communication, Bachelor's of Engineering Degree
DOI: 10.1109/ELINFOCOM.2014.6914362. in Computer Science and
Engineering.
[19] Min Hu, Yaona Zheng, Fuji Ren, He Jiang, Age
estimation and gender classification of facial images
based on Local Directional Pattern, 2014 IEEE 3rd
International Conference on Cloud Computing and Harshitha M N, Student of
Intelligence Systems, DOI: Maharaja Institute Of Technology
10.1109/CCIS.2014.7175711. Thandavapura, Mysore. Pursuing
Bachelor's of Engineering Degree
[20] Chi Xu, Yasushi Makihara; Ruochen Liao, Hirotaka Computer Science and Engineering.
Niitsuma; Xiang Li, Yasushi Yagi, Jianfeng Lu, Real-
Time Gait-Based Age Estimation and Gender
Classification from a Single Image, 2021 IEEE Winter
Conference on Applications of Computer Vision
(WACV).

© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 281