Scribd
Upload
4 views13 pages

Deep Fake Deection

The document outlines a project focused on developing a system for detecting deepfakes using advanced AI techniques, including convolutional neural networks and GANs. It aims to enhance the accuracy of deepfake detection to combat misinformation and protect privacy. The project also reviews existing literature on deepfake detection methods and their effectiveness.

Uploaded by

CSBS17Mintu Mondal
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
4 views13 pages

Deep Fake Deection

The document outlines a project focused on developing a system for detecting deepfakes using advanced AI techniques, including convolutional neural networks and GANs. It aims to enhance the accuracy of deepfake detection to combat misinformation and protect privacy. The project also reviews existing literature on deepfake detection methods and their effectiveness.

Uploaded by

CSBS17Mintu Mondal
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 13

DEEP FAKE DETECTION

BY:
Mintu Mondal (14231121023)
Tanusree Paul (14231122042)
Ayantika Paul (14231122044)
Pritha Samanta (14231121025 )
EXPLORE NOW
Chandan koley (14231121014)
Prince Kumar Thakur (14231121013)

Guided by:
Mr. Biplab Mandal
INTRODUCTION
Advances in AI have led to the creation of synthetic media that can
manipulate realistic visual content.

This project aims to develop a system to detect deepfakes and


differentiate between real and manipulated media.

The system will analyze visual content in real-time to identify


inconsistencies and signs of manipulation.

It is designed to enhance the accuracy of deepfake detection,


addressing growing concerns about misinformation and privacy.

Our goal is to create an effective tool to combat digital


disinformation, ensure the integrity of visual media, and handle
real-time data efficiently.
PROBLEM AREA IDENTIFICATION

Rapid Growth of Erosion of Human


Deepfakes Trust Limitation

Misinformation Lack of Effective


and Misuse Solutions
OBJECTIVE
Identify both spatial (image)
01
and generative (fake creation)
flaws.

Ensure reliable and efficient


02
deepfake detection.

03
WORK FLOW

Dataset Used

FaceForensics ++

Deepfake Detection
Challenge(DFDC)
PROPOSED SOLUTION OUTLINE

01 Convolutional neural network

02 Xception Model(Base Line Model)

03
03 GAN(Generative Adversal Network)

07
CNN

Layers of CNN
Convolution Layer: Extracts features
like edges and corners
Pooling Layer: Dimension Reduction
Fully Connected Layer: Combines all
extracted features to classify the input.

Limitations
Can't Detect Video Changes Well
Misses Small Details
Requires High Computing Power
Not Good with New DeepfakesEasily
Fooled by Small Changes
WORK DONE
*Results obtained from our baseline model
Xception

It uses depthwise separable convolutions to improve computational efficiency

Advantages
Better Feature Extraction
Higher Accuracy
Faster Training
Improved Efficiency
Better at Detecting Fine Details
GAN

How GANs Work Impotance of GAN in Deepfake


Generator: Creates fake media. Detection
Discriminator: Identifies if media is real or Simulation of Realistic Deepfakes
fake. Integration with Other Models
The generator improves until it fools the Adversarial Learning for Improved Performance
discriminator. Better Handling of Evolving Deepfake Methods
FIELD/LITERATURE SURVEY
Rui Zhang, Zixuan Jiang, and Changxu Sun in their paper” Two-Branch Deepfake Detection Network Based on Improved Xception” (2023) propose a
01 two-branch detection network using an improved Xception model for deepfake detection. The whole branch analyzes entire videos, while the local
branch inspects individual frames with data enhancements like Face-Cutout. Their method outperforms others in cross-dataset and cross-
definition scenarios, as shown through testing on the FF+ and Celeb-DF datasets[1].

The paper "Generation and Detection of Deepfakes using GANs and Affine Transformation" by Dr. J. Vijaya et al. (2023) presents a method that uses GANs and
affine transformations to generate deepfake videos by combining target images with driving videos. It also proposes a classification model to detect the
02 authenticity of these videos, leveraging facial recognition and forgery detection. The approach aims to improve security and protect against identity theft and
cyber-crime while exploring applications in entertainment and education[2].

Ahmed Hatem Soudy et al. (2024) propose a deep learning-based deepfake detection system using CNNs and Vision Transformers. The system processes face
03 images and uses a majority voting approach for predictions, achieving 97% accuracy with CNNs. It outperforms previous models, particularly in detecting
deepfakes on social media[3].

Sabareshwar D et al. (2024) propose a lightweight CNN architecture for efficient deepfake detection in low-resolution images in the frequency domain. The model
04 achieves similar accuracy to traditional networks while reducing parameter requirements by 92%, making it ideal for memory and power-constrained
environments like mobile devices[4].
REFERENCES

01 R. Zhang, Z. Jiang, and C. Sun, "Two-Branch Deepfake Detection Network Based


on Improved Xception," 2023.

02 J. Vijaya, A. A. Kazi, K. G. Mishra, and A. Praveen, "Generation and detection of


deepfakes using GANs and affine transformation," 2023.

03 A. H. Soudy, O. Sayed, H. Tag-Elser, R. Ragab, and T. Mostafa, "Deepfake


detection using convolutional vision transformers and CNNs," 2024.

04 S. D, R. S, S. Ravi, D. V. M, and P. M. U, "A lightweight CNN for efficient deepfake


detection of low-resolution images in frequency domain," 2024
THANK YOU
02

You might also like