Scribd
Upload
39 views13 pages

Machine Learning Techniques in Adversarial Image Forensic

The document discusses machine learning techniques for detecting deepfake and adversarial images. It describes two modules, the first for deepfake detection using custom CNN models and transfer learning models. The second module detects adversarial images using ResNet and DenseNet against FGSM and PGD attacks.

Uploaded by

rogerkali007
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
39 views13 pages

Machine Learning Techniques in Adversarial Image Forensic

The document discusses machine learning techniques for detecting deepfake and adversarial images. It describes two modules, the first for deepfake detection using custom CNN models and transfer learning models. The second module detects adversarial images using ResNet and DenseNet against FGSM and PGD attacks.

Uploaded by

rogerkali007
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 13

Machine Learning Techniques in

Adversarial Image Forensic


1)Abdullah Zahid Khan
2)Prof. Satish S. Kumbhar
3)Dr. Puneet Bakshi
8th International Conference On
Engineering Research and Innovations-(ICERI-2023) | Virtual Conference
Biography:-
1.) Abdullah Zahid Khan pursuing his Master of Technology degree from the College of
Engineering Pune (COEP), complementing his educational journey that commenced
with a Bachelor's degree from G.H. Raisoni College of Engineering and a Diploma from
PDEAs Institute of Technology. Specializing in Deep Learning, Image Forensics, and
Data Analysis.
2.) Professor Satish S. Kumbhar is a seasoned educator at COEP, boasting a
remarkable teaching tenure spanning over three decades. His expertise lies in the
domains of networking, security, and machine learning.
3.) Dr. Puneet Bakshi is currently serving as a Joint Director at C-DAC Pune, an Indian
autonomous scientific society operating under the Ministry of Electronics and
Information Technology. C-DAC Pune is renowned for its cutting-edge research and
development initiatives.

8th International Conference On


Engineering Research and Innovations-(ICERI-2023) | Virtual Conference
Introduction:-
 Significance of Image Forensic Techniques in legal investigations.
 Vulnerabilities of existing Machine Learning techniques in Image
Forensics.
 Necessity for tailored strategies.

8th International Conference On


Engineering Research and Innovations-(ICERI-2023) | Virtual Conference
 Adversarial Images: Images that have pixels purposely and
intentionally manipulated to confuse and deceive models.

8th International Conference On


Engineering Research and Innovations-(ICERI-2023) | Virtual Conference
 Deep Fakes: Deepfake technology is a method for manipulating
videos, images, audios utilizing powerful computers and deep
learning.
 It is used to generate fake news and commit financial fraud among
other wrong doings.

8th International Conference On


Engineering Research and Innovations-(ICERI-2023) | Virtual Conference
Module One - Deepfake Image Detection
 Diverse CDAC dataset for training and evaluation.
 StyleGan Generated 70000 Deepfakes from 70000 Real Images

8th International Conference On


Engineering Research and Innovations-(ICERI-2023) | Virtual Conference
Results: Module One
Model Accuracy Precision Recall
Custom CNN Model 0.95 0.92 0.99
Custom CNN Augmented 0.91 0.86 0.98
VGG Face 0.94 0.92 0.95
DenseNet 0.93 0.98 0.88
DenseNet Augmented 0.87 0.91 0.83
DenseNet Grayscale 0.93 0.98 0.89

8th International Conference On


Engineering Research and Innovations-(ICERI-2023) | Virtual Conference
Module Two - Adversarial Image Detection
 Utilized the CIFAR10 dataset for meticulous implementation and
evaluation.
 FGSM Attacks
 Alter images using loss gradients.
 Targeted FGSM Attacks
 Manipulate images for erroneous classification
 PGD Attacks
 Iterative perturbation application.

8th International Conference On


Engineering Research and Innovations-(ICERI-2023) | Virtual Conference
Module Two - Adversarial Image Detection

8th International Conference On


Engineering Research and Innovations-(ICERI-2023) | Virtual Conference
Results: Module Two
Metric ResNet DenseNet
Accuracy of PGD 91 90
PGD Attack (3 iterations) 90.4 89.22
PGD Attack (7 iterations) 89.58 88.12
PGD Attack (12 iterations) 88.89 86.68
Accuracy of FGSM 89 85.97
FGSM Attack (ϵ = 0.0157) 89.22 86.66
FGSM Attack (ϵ = 0.0314) 88.73 86.15
FGSM Attack (ϵ = 0.0471) 88.13 85.52

8th International Conference On


Engineering Research and Innovations-(ICERI-2023) | Virtual Conference
Conclusion:

8th International Conference On


Engineering Research and Innovations-(ICERI-2023) | Virtual Conference
Appendix:

8th International Conference On


Engineering Research and Innovations-(ICERI-2023) | Virtual Conference
Thanking you…!!
• All The Best for
the best in you…!!

8th International Conference On


Engineering Research and Innovations-(ICERI-2023) | Virtual Conference

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy