FINAL PROJECT SYNOPSIS

ON
“AGE AND GENDER DETECTION”
Submitted in partial fulfillment of the requirements for the award of degree of

BACHELOR OF

ENGINEERING IN
COMPUTER SCIENCE & ENGINEERING(AIML)

Submitted to: ARUN MITTAL

Submitted By:
ANKIT KUMAR(20BCS6801)
SAIKAT DAS(20BCS6818)
BIPANDEEP SINGH (68xx)
ABHIRAJ (20BCS68)

DEPARTMENT OF COMPUTER SCIENCE & ENGINEERING

Chandigarh University,

Gharuan
 Table of Contents

Topic Page No.

DECLARATION 1

LIST OF TABLES 2

ABSTRACT 3

INTRODUCTION 4

LITERATURE SURVEY 5

PROBLEM FORMULATION 6

RESEARCH OBJECTIVE 6

METHODOLOGY 7-10

TENTATVE CHAPTER PLAN FOR THE PROPOSED WORK 10-13

RFERENCES 14
 DECLARATION

We, student of Bachelor of Engineering in Computer Science & Engineering , 4 th Semester ,

session: JAN-JUNE 2024, Chandigarh University, hereby declare that the work presented in this
Project Report entitled “AGE AND GENDER DETECTION ” is the outcome of our own work,
is bona fide and correct to the best of our knowledge and this work has been carried out taking
care of Engineering Ethics. The work presented does not infringe any patent work and has not
been submitted to any other university or anywhere else for the award of any degree or any
professional diploma.

Students details and Signature

ANKIT KUMAR(20BCS6801)
BIPANDEEP(20BCS68)
SAIKAT(20BCS6818)
ABHIRAJ(20BCS68)

APPROVED & GUIDED BY:

To our Project In charge “ARUN MITTAL”.

1)
 LIST OF TABLES

S NO. FIGURE NAME

1. USE CASE DIAGRAM

2. SEQUENCE DIAGRAM

3.  Software - Microsoft Windows 98/2000/NT with

service pack 6 / XP with service pack 2/ Windows 7 with
service pack 2.
 Python

4.  Hardware-
 Processor (CPU) with 2 gigahertz (GHz)
frequency or above
 A minimum of 2 GB of RAM
 Monitor Resolution 1024 X 768 or higher
 A minimum of 20 GB of available
space on the hard disk .Internet
Connection Broadband (high-
speed) Internet connection with a
speed of 4 Mbps.

2)
 ABSTRACT

Gender classification is a binary classification issue that entails determining

whether a set of face photographs is female or male. Although there are a
variety of ways for gender categorization, including gait, iris, hand shape, and
hair, the most common method is based on facial traits. Novel approaches are
suggested in this study to accomplish the goals of (1) gender categorization and
(2) age detection in a three-step procedure. To begin, the input image collection
is pre-processed using noise reduction, histogram equalisation, and size
normalisation, followed by face detection. The second step is to extract
features from a face picture. Finally, tests were conducted on several picture
sets including an equal proportion of male and female using a binary SVM
classifier to classify the data set into two categories, male or female, in order to
evaluate the performance of the proposed approach. To fulfil the second aim, a
multi-class SVM was used to produce three classes: kid, adult, and elderly. The
age of the input photographs is determined and categorises them into one of
three groups.
 3)

INTRODUCTION
The face is the most distinguishing feature of a human being. The many facial expressions may
be used to discern and understand a person's varied moods. Because picture capture and
prediction are easier to execute on the face than on other biometric qualities, it is regarded the
most acceptable biometric attribute. Faces are typically classified as semi-rigid, semi-flexible,
culturally relevant, and a component of our individual being, necessitating effective face
recognition and classification computer systems.

Gender recognition, which uses frontal face photographs, has found applications in
authentication, search engine accuracy, demographic data gathering, human computer
interaction, access control, and surveillance. It may also be used as an indexing strategy for
automatic facial recognition to minimise the search space. The goal of this research is to do
gender categorization of human facial images using features retrieved from the input picture
collection. The 'lip,' which has been discovered and extracted from a human face picture using
the Region of Interest (ROI) concept, is the feature on which gender recognition has been done.

The derived feature is then fed into a Support Vector Machine Classifier (SVM), which is trained
on a data set that includes both male and female lip pictures. Aside from that, the aforementioned
approach was applied to a group photograph from which individual male/ female photos were
recovered. Then a similar Feature Extraction and Gender Classification technique is carried out.

Finally, an effort was made to do age detection on a combined collection of photographs, and the
 images were classed as 'child', 'adult', or 'old' depending on their age. Multi-Class SVM is used to
do this classification.

1.2 Software and Hardware tools required for Project-

Software:

System Requirements :

 Microsoft Windows 98/2000/NT with service pack 6 / XP with service pack 2/ Windows
7 with service pack 2.
 Python

Hardware:

 Processor (CPU) with 2 gigahertz (GHz) frequency or above

 A minimum of 2 GB of RAM
 Monitor Resolution 1024 X 768 or higher
 A minimum of 20 GB of available space on the hard disk .Internet
Connection Broadband (high-speed) Internet connection with a
speed of 4 Mbps.

4)
 LITERATURE SURVEY

Existing System:

Detecting a speaker's age and gender from a brief verbal utterance is a difficult subject that has lately
received a lot of attention. Metze et al. (2007) compared four approaches for age and gender
recognition from telephone speech: a parallel phoneme recognizer system that compared the Viterbi
decoding scores for each category-specific phoneme recognizer, a system that combined several
prosodic features using dynamic Bayesian networks, a system based solely on linear prediction
analysis, and a GMM system based onMFCCs. According to Metzeet al. (2007), the parallel phone
recognizer system performs as well as human listeners on lengthy utterances but suffers on short
utterances, whereas the prosodic features-based system performs better. Gender is perceived not just
by looking at the face, but also by looking at the hair, clothes, and skin tone [15, 6], stride [14], and
the entire body [6, 1]. We discuss relevant studies on gender prediction using solely face photos in the
sections below. The first is proposed by Moghaddam et al. [20], who use a Support Vector Machine
(SVM) to classify gender using thumbnail face photos. Baluja et al.[3] introduced the second, which
used the Adaboost algorithm to estimate gender. Yang et al.[30] employed LBP histogram features for
gender feature representation, and the Adaboost method to train the optimal local features for
classification, due to the prevalence of Local Binary Patterns (LBP) in face recognition applications
[2]. Face photos were used in experiments to determine age, gender, and ethnicity. In [25], a similar
strategy was presented. For gender categorization, several local descriptors have been used. In this
study, we employ LBP in conjunction with SVM to determine gender from face photos as a baseline
for appearance-based approaches.Although geometry characteristics were employed as a priori
information to enhance classification performance in previous work [22, 29], none of the
aforementioned systems, unlike ours, relied explicitly and completely on face metrology as a means
of gender categorization. Perhaps our approach is more closely connected to prior work on face
identification using geometric characteristics by Shi et al. [23, 24], in which they employed ratio
features calculated from a few anatomical landmarks.However, we take a closer look at how explicit
face geometry is used to solve the gender categorization problem. We only employ metrological data
based on landmarks, which may or may not have biological significance.Instead of using holistic
information from all landmarks, we employ local information from independent landmarks.

Proposed System:
This section presents the proposed deeply learned classifiers for age group and gender classification
of unfiltered real-life face images.
The approach requires image preprocessing (face detection, landmark detection, and face alignment)
stage that preprocess and prepare the face images before they are input into the proposed network.
Therefore, our solution is divided into three major steps: image preprocessing, features learning, and
classification itself.

5)
 PROBLEM FORMULATION
The project is totally build at administrative end and thus only
the administrator is guaranteed to access. To manage the
details of the age and gender. The purpose of the project is to
build an application program to reduce the manual work for
managing the age an gender detection. It tracks all the details
about the person age and gender.
REEARCH OBJECTIVE
We tackled the classification of age group and gender of unfiltered
real-world face images. We posed the task as a multiclass
classification problem and, as such, train the model with a
classification-based loss function as training targets. Our proposed
model is originally pretrained on age and gender labelled large-scale
dataset, whose images are obtained directly from the website with
some degree of variability and then fine-tuned on another large-scale
facial aging dataset with age and gender annotations. Finally, we use
the original dataset to fine-tune this model. The robust image
preprocessing algorithm, handles some of the variability observed in
typical unfiltered real-world faces, and this confirms the model
applicability for age group and gender classification in-the-wild.
Finally, we investigate the classification accuracy on Adience dataset
for age and gender; our proposed method achieves the state-of-the-art
performance, in both age group and gender classification, significantly
outperforming the existing models. For future works, we will consider
a deeper architecture and a more robust image processing algorithm
for exact age estimation. Also, the apparent age estimation of human’s
face will be interesting research to investigate in the future.
6)
 METHODOLOGY

Technology to be used

Application Scripts

Predict.py
Contains class Predictor. It is responsible for loading the models,
running prediction on the input and drawing the result on the input
image. Additionally it is responsible for parsing command line
arguments and processing the input from command line interface. If the
application is used in conjunction with person tracking, Predictor class is
responsible for computing the average value of all previous prediction
for one person.

face_detect.py
Contains class Detector. It is responsible for detecting faces on the input
image. If the face filtering is switched on then the Detector class will
pass the detected faces to face_filter . py. Detector returns detected faces
and the coordinates of their bounding boxes on the original image.

face_filter.py
Face Filter class is responsible for finding the center of each eye and
nose line using facial landmarks. It then computes the ratio of distances
1 Application Scripts In this Section we wil l detail the important Python
scripts of the application.

7)
Third Party Libraries

NumPy

NumPy is a Python library for mathematics. It provides support for large

multidimensional matrices. It also includes various linear algebra
functions. It is licensed under BSD 3-clause license. NumPy array is a
primary mode of storing and manipulating with images. Pandas Pandas
is a Python data analysis library. It contains data structures and functions
for manipulating with tables. It is released under BSD 3-clause license.
We use pandas during data preprocessing to parse dataset labels.

8)
In this section, we'll talk about the approach that's been used to
divide a group of photos into male and female categories.

The block diagram of the proposed model is given below:-

9)
Algorithm:
Step 1: Read the Male and Female Image Set
Step 2: Grayscale each of the input images.
Step 3: Perform feature extraction from each image using the ROI method.
Step 3.1: Perform the instructions below for each image.
Step 3.1.1: Remove the 'lip' from each picture.
Step 3.1.2: Reduce the size of the retrieved picture from two dimensions to one dimension.
Step 3.1.3: Assign a class label to each image. Give the female picture a +1 and the male image a -1.
Step 3.1.4: Combine the retrieved photos and the class label into a Feature Vector.
end
Step 4: Shuffle the matrix of Feature Vectors.
Step 5: Validate the matrix with cross-validation and create the train and test data sets.

TENTATVE CHAPTER PLAN FOR THE PROPOSED WORK

DIFFERENT FACE RECOGNITION APPROACHES:

Face recognition is usually approached in one of two ways: geometric (feature-based) or photometric
(view based). Face recognition has piqued the curiosity of researchers.Many distinct algorithms were
created as a result, three of which have been extensively researched in the face recognition literature.

1. Geometric: Is based on the spatial layout of face features, or the geometrical link between facial
landmarks. That is, the key geometrical characteristics of the face, such as the eyes, nose, and mouth,
are found first, and then faces are categorised using geometrical distances and angles between
features.

2. Photometric stereo: This technique is used to reconstruct an object's form from a series of photos
collected under various lighting conditions. A gradient map, which is made up of an array of surface
normals, defines the form of the recovered object.

10)
 FACE DETECTION:

Face detection involves separating image windows into two classes; one containing faces (tarning the
background (clutter). It is difficult because although commonalities exist between faces, they can vary
considerably in terms of age, skin colour and facial expression. The problem is further complicated by
differing lighting conditions, image qualities and geometries, as well as the possibility of partial
occlusion and disguise. An ideal face detector would therefore be able to detect the presence of any
face under any set of lighting conditions, upon any background.

11)
MODEL APPROACH

INTERFACE REQUIREMENTS
SOFTWARE INTERFACE
 Windows 8/10.
 4 gb ram / 256gb hdd
 MySQL
 Html
 Javascript & PHP

Several materials are used to built this project . Certain software interfaces are used
like window 8/10 , 4gb ram/ 255gb hdd , MySQL . Technologies are also used like
Javascript for development of the project. MySQL is used for free open source
database to store the data .

12)
 CONCLUSION

We implemented an application for age and gender prediction. The application can be used both as a
Python module imported into another project and as a standalone application with command line
interface. The application provides two different face detection method. One is based on computing
histogram of oriented gradients, the second is based on cascade of convolutional networks.
Additionally the application provides a way to filter faces which are turned away from the camera.
We also enable the usage of our application in conjunction with person detection tools by accepting a
person id parameter and computing the average result.
The defined procedures can be enhanced in the future by including the specifications listed below into
the execution of the suggested mechanisms
With a larger number of face picture data sets, the Gender Classification and Age Detection
algorithms may be used. The output's accuracy will improve as a result of this.
Linear Kernel was used to train and evaluate the proposed models on data sets. However, additional
kernels of the SVM Classifier, such as the 'rbf kernel' and the 'quadratic kernel,' can be used to do a
similar assessment.
The retrieved feature' lip' is used to achieve gender classification. The same technique may be used to
detect additional features such as eyes, noses, or a combination of features.
We also trained one age model and one gender model using integral image as an additional colour
channel. This did not improve the performance and the results on most datasets were worse that the
pretrained models. It is possible that training with integral images requires a different neural network
architecture or different set of hyperparameters.

Possible Future Research

There are many possibilities in age and gender estimation research. A n immediate idea would be to
look more deeply into training models with integral images as additional colour channels using more
varied neural network architectures. Another idea would be to use more varied neural network
architecture specifically for gender prediction. Many current tools use the same architecture for both
age and gender prediction.

13)
 REFERENCES

1.Golomb BA, Lawrence DT, Sejnowski TJ. Sexnet: A neural network identifies sex from human
faces. Advance in Neural Information Processing Systems. 1990;572-577.
2. Wen-Sheng Chu, Chun-Rong Huang, ChuSong Chen. Identifying gender from unaligned facial
images by set classification. In Proc. IEEE 20th
International Conference on Pattern Recognition. 2010;2636–2639.
3. Ameneh Shobeirinejad, Yongsheng Gao.Gender classification using interlaced derivative patterns.
In Proc. IEEE International Conference on Pattern Recognition. 2010;1509-1512.
4. Li LU, Ziyi XU, Pengfei SHI. Gender classification of facial images based on multiple facial
regions. World Congress on Computer Science and Information Engineering. 2009;48-52.
5. Gregory Shakhnarovich Paul A. Viola Baback Moghaddam. A unified learning framework for real
time face detection and classification. In Proc. IEEE International Conference on Automatic Face and
Gesture Recognition; 2002.
6. Han X, Ugail H, Palmar I. Gender classification based on 3D face geometry features using SVM. In
Proc. IEEE International Conference on Cyber World.2009;114-118.
7. Wang Yiding, Zhang Ning. Gender classification based on enhanced PCASIFT facial features. In
Proc. IEEE international conference on information science and engineering. 2009;1262-1265.
8. Ravi S, Wilson S. Face detection with facial features and gender classification based on support
vector machine. In Proc. IEEE International Conference on Computational Intelligence and
Computing Research. 2010;23-28.
9. Available: http://en.wikipedia.org/wiki/Histogram_equalization
10. Available: http://in.mathworks.com/help/images/noise-removal.html
11. Available: http://www.cs.utexas.edu/~grauman/courses/378/handouts/moghaddam2000.pdf.
12. Available: http://www.csie.ntu.edu.tw/~cjlin/papers/multisvm.pdf
13. Available: http://arxiv.org/ftp/arxiv/papers/0802/0802.2411.pdf
14. Available: https://hal.archivesouvertes.fr/inria-00103955/document
15. Available: http://www.csie.ntu.edu.tw/~cjlin/papers/guide/guide.pdf
16. Available: http://www.cs.cmu.edu/~schneide/tut5/node42.html

14)