GOVERNMENT COLLEGE

UNIVERSITY FAISALABAD

Project Proposal

For

Fake Face Detection

By

Muhammad Dawood 2019-GCUF-077140

Hafiza Kulsoom Sarwar 2019-GCUF-077141
Tasleem Bibi 2019-GCUF-077142

Supervisor
Umer Yasin

Bachelor of Science in Computer Science (2019-2023)

SCOPE DOCUMENT REVISION HISTORY

No. Comment Action

Supervisor Signature

Date:
Abstract
Technology improvements have made it possible to seamlessly represent human faces in fake media
by using the widespread availability of fake images. Attacks on the targets may be carried out using
these fictitious faces. Open-source software and a wide range of commercial programs make it
possible to create fake images of specific target subjects in a variety of methods. To benchmark the
detection accuracy, we assess the generalizability of the fake face detection techniques. To this aim,
we have gathered a new database of images that come from many sources of digitally made fakes,
including Computer Graphics Image (CGI) production and many tampering-based approaches.
Additionally, we have added pictures from the widely popular Swap-Face app, which is accessible on
most smartphones. To determine the applicability of detection methods, extensive experiments are
conducted utilizing deep learning-based detection methods. We try to determine whether the present
faked face detection techniques are generalizable through the set of evaluations.

Table of Content
s

1. INTRODUCTION..........................................................................................................................1
2. PROBLEM STATEMENT.............................................................................................................1
3. PROBLEM SOLUTION FOR PROPOSED SYSTEM..................................................................1
4. ADVANTAGES/BENEFITS OF PROPOSED SYSTEM.............................................................1
5. RELATED SYSTEM ANALYSIS/LITERATURE REVIEW....................................................2
6. SCOPE............................................................................................................................................2
7. MODULES.....................................................................................................................................2
7.1 Face Processing:......................................................................................................................2
7.2 Deep feature extraction:..............................................................................................................2
7.3 Face matching:.............................................................................................................................2
8. SYSTEM LIMITATION/CONSTRAINTS....................................................................................3
9. TOOLS AND TECHNOLOGY......................................................................................................3
10. FUNCTIONAL REQUIREMENTS...............................................................................................4
11. PROJECT STAKEHOLDERS AND ROLES................................................................................4
12. DATA GATHERING APPROACH...............................................................................................4
13. GANTT CHART............................................................................................................................5
14. CONCLUSION...............................................................................................................................5
15. REFERENCES...............................................................................................................................6
 1. INTRODUCTION

It can be challenging to distinguish between authentic and fake information in the age of
social media, whether it's news or a picture or video of a politician, celebrity, or other figures.
Additionally, a lot of phony or altered faces and films are being produced, which are more
difficult to spot using conventional tools or techniques. Therefore, it is possible to easily
distinguish between real and fake photos, faces, and videos by using Deep Learning, a subset
of Machine Learning. We will determine whether the faces in the provided dataset are real or
phony.
2. PROBLEM STATEMENT
Deep learning algorithms have the ability to produce fake photos and movies that look real to humans.
Therefore, the development of systems that can instantly identify and evaluate the integrity of digital
visual media is essential. It can be challenging to distinguish between authentic and fake information
in the age of social media, whether it's news or a picture or video of a politician, celebrity, or other
figure. Additionally, a lot of phony or altered faces and films are being produced, which are more
difficult to spot using conventional tools or techniques.

3. PROBLEM SOLUTION FOR PROPOSED SYSTEM

Deep learning, a branch of machine learning, can be used to quickly determine if photos, videos, or
faces are authentic or phony. We will determine whether the faces in the provided dataset are real or
phony. The software examines images to determine with what confidence the material was likely
created artificially.

4. ADVANTAGES/BENEFITS OF PROPOSED SYSTEM

 Improved security. Face detection enhances surveillance and aids in the capture of terrorists
and criminals. Since there is nothing for hackers to take or alter, like passwords, personal
security is also increased.
 Easy to integrate. The majority of solutions for face detection and facial recognition are
compatible with the vast majority of security software, making integration simple.
 Automated identification. Identification used to be done manually, which was frequently
inaccurate. Automating the identification process with face detection increases accuracy and
reduces processing time.

1
 5. LITERATURE REVIEW

System Review Proposed Project Solution

[IMVOP 2018] Detection of This system will be more
Deep fake Image Detection Accurate and support
normal range devices.

6. SCOPE
 Face recognition is useful in a variety of industries. According to the findings of our study, the
following may be said about the scope:
 Room for improving accuracy: Multiclass classifier accuracy can be increased. For a huge
database, many alternative methodologies may be used and contrasted to provide superior
accuracy results.
We may apply it in a variety of security-related domains to gain improved accuracy.

7. MODULES

This system will have the following modules:

7.1 Face Processing:
Face normalization to frontal view is accomplished through face processing in order to improve
accuracy.
7.2 Deep feature extraction:
To identify particular people with the proper loss functions, deep feature extraction uses a deep
learning network architecture that has been trained on large-scale face data sets.
7.3 Face matching:
These face feature vectors will be used by the face matching module to build an appropriate model for
calculating the distance between the face feature vectors for face recognition and verification using
deep learning.

2
8. SYSTEM LIMITATION/CONSTRAINTS

 Knowledge-based, or rule-based methods, describe a face based on rules. The challenge of this
approach is the difficulty of coming up with well-defined rules.

 Feature invariant methods -- which use features such as a person's eyes or nose to detect a face
can be negatively affected by noise and light.

 Template-matching methods are based on comparing images with standard face patterns or
features that have been stored previously and correlating the two to detect a face.
Unfortunately these methods do not address variations in pose, scale and shape.

9. TOOLS AND TECHNOLOGY

Tools Version Rationale

MS Visual Studio 2022 IDE
MS Word 2021 pro Documentation
Tools MS Power Point 2021 pro Presentation
And Google Colab For python code
Technologies Numpy Library for the
phython
Tenser Flow Library
Keras API
Technology Version Rationale
Python 2.7 Programming
language
Open CV 2.0 Computer Vision
Deep learning Machine learning
technique

3
10.FUNCTIONAL REQUIREMENTS

The following are the functional requirements of the project:

1. The tool based application/software must download the given Dataset that contains the database of
real and fake images.
2. The model must be any CNN model that contains hidden layers for the fake face detection.
3. Whenever, any face is given as an input into the detection model, it identifies as real or fake as
output.
4. The model must be able to recognize the fake faces generated by any of the apps like
FaceApp, FaceSwap, Wombo etc.

11. PROJECT STAKEHOLDERS AND ROLES

Project
COMSATS University, Islamabad
Sponsor
Stakeholder • Muhammad Dawood.
• Final Year Project Committee

12. DATA GATHERING APPROACH

Techniques that are used for collecting requirements are as follows:
Dataset:
https://www.kaggle.com/xhlulu/140k-real-and-fake-faces
Deep Learning for Face Recognition:
https://machinelearningmastery.com/introduction-to-deep-learning-for-face-recognition/
Convolutional Neural Network:
https://towardsdatascience.com/convolutional-neural-networks-explained-9cc5188c4939

4
13. GANTT CHART

14. CONCLUSION

We suggested developing a face recognition system that is high performing, scalable, flexible, and
affordable. We break down the suggested strategy into a number of tiny side projects. Convolutional
and neural networks were first researched. We constructed the network, which would train the neural
network based on similarities, based on the idea of deep learning. Following a review and comparison
of the open source data sets available, we selected the ORL dataset and GPU-trained the model. A
human face image will be taken by the model, who will then extract it as a vector. Then it is
determined if two faces on distinct pictures belong to the same person by comparing the distance
between the vectors. After that, we carried out research, made comparisons, designed a system, and
built it to function with the neural network model. The architecture of the system is client-server. On
the server side, GPU is employed to deliver great performance. In order to make the system adaptable
and scalable, the primary components were also decoupled. To boost the concurrency of the system,
we made use of Node.JS's and asynchronies features. Since the entire system is modular, it can be
5
used to many domains, which lowers the cost of development.

15. REFERENCES
[1] "cookbook.fortinet.com," 10 10 2018. [Online]. Available: https://cookbook.fortinet.com/face-
recognition-configuration-in-forticentral/. [Accessed 10 10 2018].
[2] M. J. Paul Viola, "Robust Real-time Object Detection," International Journal of Computer Vision,
pp. 137-154, 2004.
[3] H. A. C. H. a. S. L. Bo Wu, "Fast rotation invariant multi-view face detection based on real
Adaboost," Sixth IEEE International Conference on Automatic Face and Gesture Recognition, pp. 79-
84, 2004.
[4] H. A. Y. T. S. L. Yuan Li, "Tracking in Low Frame Rate Video: A Cascade Particle Filter with
Discriminative Observers of Different Life Spans," IEEE Transactions on Pattern Analysis and
Machine Intelligence, pp. 1728-1740, 2008.