Scribd
Upload
59 views21 pages

Detection of Fake and Fraudulent Faces Via Neural Memory Networks

The document proposes a Hierarchical Attention Memory Network (HAMN) for detecting fake and fraudulent faces. HAMN is inspired by human social cognition and uses neural memory networks to reason about faces and anticipate future representations, allowing it to generalize to unseen manipulation types. It outperforms CNNs that overfit to specific manipulations. The network captures patch-level and image-level semantics and propagates knowledge through memory hierarchies to predict temporal face evolution and discriminate real from fake faces.

Uploaded by

rajkumarmtech
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
59 views21 pages

Detection of Fake and Fraudulent Faces Via Neural Memory Networks

The document proposes a Hierarchical Attention Memory Network (HAMN) for detecting fake and fraudulent faces. HAMN is inspired by human social cognition and uses neural memory networks to reason about faces and anticipate future representations, allowing it to generalize to unseen manipulation types. It outperforms CNNs that overfit to specific manipulations. The network captures patch-level and image-level semantics and propagates knowledge through memory hierarchies to predict temporal face evolution and discriminate real from fake faces.

Uploaded by

rajkumarmtech
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/ 21

DETECTION OF FAKE AND FRAUDULENT FACES VIA

NEURAL MEMORY NETWORKS

1
ABSTRACT

Advances in computer vision have brought us to the point where we have the
ability to synthesise realistic fake content. Such approaches are seen as a source of
disinformation and mistrust, and pose serious concerns to governments around the
world. Convolutional Neural Networks (CNNs) demonstrate encouraging results
when detecting fake images that arise from the specific type of manipulation they
are trained on. However, this success has not transitioned to unseen manipulation
types, resulting in a significant gap in the line-of-defense. We propose a
Hierarchical Attention Memory Network (HAMN), motivated by the social
cognition processes of the human brain, for the detection of fake faces. Through
visual cues and by utilising knowledge stored in neural memories, we allow the
network to reason about the perceived face and anticipate it’s future semantic
embeddings. This renders a generalisable face tampering detection framework.
Experimental results demonstrate the proposed approach achieves superior
performance for fake and fraudulent face detection.
PROBLEM DEFINITION

• There exists a critical need for automated systems which capture the
fundamentals of video and image content tampering, and have the ability
to generalize to different manipulation attacks .
INTRODUCTION
• social media becoming a primary source of information for many people, and
with more than 100 million hours of video content watched daily , fake news
has quickly risen as a significant threat to society and democracy
• It is observed that the underlying neural network can quickly overfit to a
specific artefact left.
• a deep learning framework to detect fake and fraudulent faces in images and
video which is motivated by the social perception and social cognition
processes in the human brain.
• Advances in computer vision have brought us to the point where we have the
ability to synthesise realistic fake content. Such approaches are seen as a source
of disinformation and mistrust, and pose serious concerns to governments
around the world.
EXISTING SYSTEM

• A naive way to design such a system would be to use the entire temporal
structure of the given video from the modeled dataset and sequentially map
the temporal evolution of the facial expressions/attributes. However, this
limits applications of the system to videos only.
• In contrast, a system that operates at a frame-level is more flexible.
Furthermore, we observe that the majority of existing state-ofthe-art
frameworks for recognizing fake faces operate at the frame-level.
• Hence, to allow direct comparisons we design a framework that only
utilizes a single frame as the input to the system. It should be noted that the
proposed system requires image pairs (i.e current and future frames) only
for the training process.
DISADVANTAGE

• Considering the performance of the memory architectures, the worst


performance observed is for the NTM model highlighting the inadequacy
of a single level attention in the memory retrieval mechanism.
• A slight performance boost is observed through the introduction of a tree-
structured memory, however, due to the mixing of semantics between
different embedding’s that reside in the memory, this method fails to
effectively learn to discriminate fake images from real.
• The generation of a sequential representation from images is common
within the machine learning community.
• For instance in the authors utilize a similar GRU layer to summarize the
input features from the input image. Furthermore, in a similar unstacking
process is utilized in order to represent relationships among image patches
PROPOSED SYSTEM

• • We propose a method to detect tampered face images that is inspired by


the human social cognition process, and use joint learning to predict the
future face state, supporting the primary tampered media detection task.
• • We propose a novel neural memory architecture for hierarchal semantic
embedding propagation and prediction.
• We introduce an adversarial learning paradigm to train the memory
architecture where the model learns a task specific loss for both fake face
classification and future semantic generation tasks. • We perform extensive
evaluations on the proposed network using multiple public benchmarks
where we outperform baseline methods by a substantial margin, especially
for unseen tampering categories.
.
ADVANTAGES

• The main advantage of this method is the transferability of its learned


representation across different, and most importantly, unseen face
manipulation methods.
• By capturing both patch level and image level semantics, and effectively
propagating the learned knowledge hierarchically through a memory
architecture, the proposed method attains its ability to accurately anticipate
the temporal evolution of the observed face, allowing it to discriminate
fake faces from real ones.
• We would like to re-iterate the notion of the uncanny valley effect, which
also shows that synthesized content has different characteristics compared
to real content. The proposed future semantic embedding prediction task
has effectively leveraged this feature of the synthesized faces for the fake
face detection task.
ARCHITECTURE
Module Descriptions

• Neural memory networks

• The main difference between the functioning of neural networks and the biological

neural network is memory. While both the human brain and neural networks have the

ability to read and write from the memory available, the brain can create/store the

memory as well. Researchers identified that this key difference is the major roadblock

for today’s AI systems to reach human-level intelligence.

• Researchers at Deep Mind aimed to build a differentiable computer, by putting together a

neural network and linking it to external memory. The neural network would act as a

CPU with a memory attached. Such differentiable computers aim to learn programs

(algorithms) from input and output data.


• Fake and fraudulent face detection

• Convolutional Neural Networks demonstrate encouraging results when


detecting fake images that arise from the specific type of manipulation they
are trained on. However, this success has not transitioned to unseen
manipulation types, resulting in a significant gap in the line-of-defense.
• We propose a Hierarchical Attention Memory Network motivated by the
social cognition processes of the human brain, for the detection of fake
faces. Through visual cues and by utilizing knowledge stored in neural
memories, we allow the network to reason about the perceived face and
anticipate it's future semantic embeddings. This renders a generalizable
face tampering detection framework.
• Image and video forensics

• Multimedia contents are generated in many different ways through the use
of consumer electronics and high-quality digital imaging devices, such as
smartphones, digital cameras, tablets, wearable sensors, and other Internet
of Things (IoT) devices. The ever-increasing convenience of image
acquisition has facilitated instant distribution and sharing of digital images
on digital social platforms, determining a great amount of the exchanged
data
SYSTEM REQUIREMENTS

HARDWARE REQUIREMENTS
SOFTWARE REQUIREMENTS

• System : Pentium IV 2.4 GHz.


• Hard Disk : 40 GB. • Operating system : - Windows 7.
• Monitor : 15 inch VGA Color. • Front End : - JAVA or JSP
• Mouse : Logitech Mouse. • Database : - SQL SERVER
• Ram : 512 MB 2008
• Keyboard : Standard • Tools :- Eclipse IDE
Keyboard
USECASE DIAGRAM
SEQUENCE DIAGRAM
COLLABORATION DIAGRAM
Activity Diagram
COMPONENT DIAGRAM
DEPLOYMENT DIAGRAM
CONCLUSION
• Motivated by the social perception and social cognition processes of
the human brain, we have presented in this paper a Hierarchical
Attention Memory Network (HAMN) architecture for the detection of
fake and fraudulent faces. The main advantage of this method is the
transferability of its learned representation across different, and most
importantly, unseen face manipulation methods.
• By capturing both patch level and image level semantics, and
effectively propagating the learned knowledge hierarchically through a
memory architecture, the proposed method attains its ability to
accurately anticipate the temporal evolution of the observed face,
allowing it to discriminate fake faces from real ones.
• Through extensive evaluations, we have demonstrated the utility of the
hierarchical modelling of the stored knowledge while preserving the
identity of those facts, and provide visual evidence of how the memory
retrieves stored facts while considering how they relate to the current
input.
THANK YOU

You might also like