PRAGATI ENGINEERING COLLEGE

(Approved by AICTE & Permanently Affiliated to JNTUK & Accredited by NBA and NAAC)
1-378, ADB Road, Surampalem, Kakinada Dist., A.P, Pin-533437.

DEPARTMENT OF COMPUTER SCIENCE & ENGINEERING

A Major Project Work Presentation on
UNMASKING DEEP FAKES : DETECTION OF MANIPULATED IMAGES
AND VIDEOS THROUGH ADVANCED DEEP LEARNING TECHNIQUES
BY
MADAPATI YOSHITHA (20A31A0510)
VELAMPALEPU J S D SATYA SAI (20A31A0527)
SIDDHANATHI SURYA SAILAJA (20A31A0522)
TALLAPUDI BALAJI (20A31A0559)
CHATRATHI SAI SURYA (20A31A0533)
Guided by: Y.MANAS KUMAR
-Assistant Professor ,CSE DEPARTMENT

A.Y : 2023-2024
 ABSTRACT

Deepfakes are digital manipulation techniques that utilize deep learning to create misleading
images and videos. Identifying deepfake images presents the greatest challenge in discerning
the original content. As deepfakes become increasingly sophisticated and widespread, they
have the potential to deceive and manipulate viewers by presenting fabricated content as
genuine. This project explores and experiments with diverse methods for detecting both fake
and genuine images and videos. The Convolutional Neural Network (CNN) algorithm,
specifically InceptionNet, is employed to discern deepfakes. A comparative analysis is
conducted, assessing various convolutional networks. The proposed model yields superior
accuracy, achieving a 98.74% success rate in identifying deepfake images and videos.
 MOTIVATION
The motivation for the project lies in the growing threat posed by deepfake technology. This
poses significant risks to various aspects of society, including politics, journalism, entertainment,
and personal relationships.
 Preserving Trust and Authenticity: Being able to reliably identify deepfakes helps
safeguard against misinformation and preserve the integrity of visual content.
 Combating Misinformation: Deepfakes have the potential to spread false narratives,
influence public opinion, and even incite unrest.
 Advancing Technology: Researching and experimenting with different methods for
detecting deepfakes contributes to the advancement of technology in this field.
Overall, the motivation for the project stems from the need to address the challenges posed by
deepfake technology and mitigate its potential negative impacts on society, individuals, and
businesses.
 EXISTING SYSTEM
Existing systems for deepfake detection often utilize Convolutional Neural Networks (CNNs)
and may include pre-trained InceptionNet models.
Diverse datasets, both for training and testing, are essential for system development. To
enhance detection accuracy, some systems incorporate facial landmarks, audio analysis, and
temporal consistency checks.
Open-source libraries and frameworks like TensorFlow and PyTorch are commonly used in
such projects.
Real-time deepfake detection can be achieved by deploying the model in a video processing
pipeline. Continuous research and development are crucial due to evolving deepfake generation
techniques.
Up-to-date resources and collaboration with experts in the field can help refineand improve
the system's performance.
 PROPOSED SYSTEM
In the proposed system for “Unmasking Deep Fakes: Detection of Manipulated Images and
Videos through Advanced Deep Learning Techniques“, we aim to overcome the limitations of
existing systems.
By combining InceptionNet with other advanced CNN architectures and employing multi-
modal analysis, the system improves accuracy in identifying deepfake content, reducing both
false positives and false negatives.
Through a diverse and extensive dataset, the system is better equipped to generalize to new and
evolving types of deepfakes, enhancing its robustness and adaptability.
The system prioritizes model interpretability and fairness, addressing ethical concerns and
biases in deepfake detection, making it more transparent and equitable.
Advantages :
1.Enhanced system accuracy 2.Real-Time processing
 TECHNICAL SPECIFICATIONS

HARDWARE REQUIREMENTS
System Intel i5 11th gen
Hard Disk 512SSD, 2GB GRAPHIC CARD

RAM 8GB

SOFTWARE REQUIREMENTS
Operating System Windows 11

Coding Language Python 3.7.5

Software Tool VS Code 22

SYSTEM ARCHITECTURE
 DESIGN MODEL

• Algorithm: CNN – InceptionNet

• Input: Facial Image and Video
• Output: Detects deepfake and results gives result as real or fake

1 Get Image and Video using CV2 (open CV module)which helps to read image.
2. It study features using CNN InceptionNet it analyzing the dimensions of the Image.
3. It study features using CNN InceptionNet by analyzing the dimenions of each frame in the video.
4. It automatically detects real or fake Image and Video.
Use Case Diagram:
Class Diagram:
Sequence Diagram:
Activity Diagram:
Component Diagram:
Deployment Diagram:
 MODULES

Module 1:
• Image selection:
In this module an image is selected as input through file selection process.

In the above screen click on ‘Submit’ button to upload image

Module 2:
• Image Detection :
In this module input image is detected whether it is real or fake.
Module: 3
• Video selection:
In this module an image is selected as input through file selection process.

In the above screen click on ‘Submit’ button to upload image

Module 4:
• Video Detection:
In this module input video is detected whether it is real or fake.
 RESULTS

Admin home page Admin login page

Admin Dashboard
Training and testing the model
Graph Analysis User Home page
Input for prediction Prediction result
Input for prediction Prediction Result
Real-time Example :

 The recent incident of deepfake involved the creation of

manipulated videos using artificial intelligence.
 It highlighted the growing concern over deepfake technology and
its potential over misuse,including the spread of misinformation and
the exploitation of individuals.
 The incident sparked the discussions about the need for stricter
regulations and better detection methods to combat the spread of
deepfakes.
 CONCLUSION

In this project work, the InceptionNet architecture has been used for identifying the fake faces.
Different types of transitions in real images with test parameters, such as the number of key
points in images, comparison rate, and performance time required for each algorithm are used.
This project shows overall accuracy for the DFDC dataset as 98.74%. This work can classify
deepfakes recordings from various resources with diverse convolutional layers. Thus, this project
contribution will inevitably help with the diminishment of fake recordings and coercion in our
society. The proposed work was completed more faster than the existing work, and the detection
of fake and real images was very effective. In the DFDC dataset, the accuracy rate of proposed
work reached 98.74%. It could be extended in the future to use different classifiers and distance
metric measures to detect deepfake face images and videos.
 REFERENCES
 [1] Zhang, K., Zhang, Z., Li, Z., & Qiao, Y. (2016). Joint face detection and alignment using

multitask cascaded convolutional networks. IEEE signal processing letters, 23(10), 1499-1503.

 [2] Mordvintsev, Alexander, Christopher Olah, and Mike Tyka. "Inceptionism: Going deeper

into neural networks." (2015).

 [3] Badale, Anuj, et al. "Deep fake detection using neural networks." 15th IEEE international

conference on advanced video and signal-based surveillance (AVSS). 2018.

 [4] Dosovitskiy, Alexey, et al. "An image is worth 16x16 words: Transformers for image
recognition at scale." arXiv preprint arXiv:2010.11929 (2020).
THANK YOU