Deepfake Detection ( FYP) By Ghansham Amrani





ABSTRACT:

Deepfake detection refers to the process of identifying and verifying the video that which is original and which one is fake. Dee fakes are computer-generated media that use artificial intelligence (AI) and machine learning techniques to create realistic depictions of people and events never actually occurred. Deep fake detection remains a challenging and ongoing research topic due to the rapidly advancing technology and the potential for deep fake creators to improve their methods and evade detection.

 

Background of Study:

The study of deepfake detection has emerged as a response to the increasing threat posed by deepfakes - AI-generated videos or images that can be used to spread misinformation and manipulate public opinion. Deepfakes have the potential to cause significant harm to individuals, organizations, and society as a whole,and detecting them is critical for mitigating their negative impacts.

PROBLEM STATEMENT:

The problem of deep fakes refers to the creation of highly realistic and convincing manipulated media, such as videos, audios, or images, using advanced AI and machine learning techniques.

The main challenge with deep fakes is their ability to deceive viewers and evade detection, which necessitates the development of effective detection techniques and policies to mitigate their harmful impact.

If a research problem exists in deep fake detection, the findings and solutions can be used in various practical applications. For example:

Development of deep fake detection tools:The research can lead to the development of more effective deep fake detection tools that can be used by social media platforms, news organizations, and other customers to identify and remove manipulated media.

Training and educationThe research can inform the development of training and educational programs for individuals and organizations to raise awareness of the risks and potential harm associated with deep fakes and to improve their ability to detect and respond to them.

Forensic investigations: The research can inform the use of forensic techniques for investigating cases involving deep fakes, such as cyber bullying, blackmail, or identity theft.

Objectives of Study:

To identify whether a given video is real or has been manipulated using deep learning techniques.

To Avoiding the spread of false information is feasible by identifying deepfakes. False information can have a severe influence on both individuals and society as a whole.

Improve media literacy:  Deepfake detection can help improve media literacy by educating individuals on how to identify deepfakes and other forms of manipulated media.

 

 

 

LITERATURE REVIEW:

In Some countries like South Korea, Israel, USA  arleady work on deepfake detection, on there sites deepfake is common nowdays, but in pakistandeepfake is not commonly that's why deepfakedetection not avialbale here, may be some  oraganization are working but not sure. China claimthat works 95% accuracy rate.

This will be more on hype technology after some years.

There has been a significant amount of work done related to deepfake detection which are given below:

Year of Publication:

Author’s Name:

No. of Citations:

Key Findings:

2020

Zhang etal

11 

Using deep learning anderror   level   analysis(ELA)   to   identifycounterfeit   features,   anew   technique   forextracting   fake   facialfeatures   has   beendeveloped   to   separateDeepFake-generatedimages from real ones.The proposed technique can obtain accurate counterfeit feature, which enables it to outperform direct detection techniques from a simplicity and efficiency perspective

2021

Masood etal

19

With Tensorflow or Keras, open-source trained models, affordable computing infrastructure, and the rapid evolution of deep-learning (DL) methods, especially Generative Adversarial Networks (GAN)Detection and generation of audio and video deepfakes, this study examined existing tools and machine learning (ML) approaches to deepfakegeneration.

2021

Hu etal

1

A  wide   range of high-quality Deepfake videos,our method has proven tobe   highly   efficient   indetecting   in-datasetpattern   matches,detecting   within-datasetpattern   matches,   anddetecting   cross-dataset pattern matches.

 

COMPARATIVE ANALYSIS:

Features/parameters of deepfake detection:

Facial and body movementsDeepfakedetection systems can analyze the facial and body movements in a video to identify any inconsistencies or anomalies that may indicate the presence of a deepfake. These movements include eye blinks, lip movements, and head movements.

Audio analysis: Deepfake detection systems can also analyze the audio in a video to detect any anomalies or inconsistencies that may indicate the presence of a deepfake. This includes changes in tone, pitch, or cadence of speech.

Deep neural networks: Many deepfakedetection systems use deep neural networks to analyze the content of a video and detect any anomalies or inconsistencies that may indicate the presence of a deepfake.

PROJECT OVERIEW:

Deepfake detection project involves the development of a robust and accurate system that can identify deepfakes in videos or images using a range of techniques and technologies. The project requires a deep understanding of machine learning, computer vision, and biometric analysis, as well as a comprehensive dataset of authentic andeepfakecontent.

 

 

 

 

 

 

PROJECT DEVELOPMENT METHODLOGY / ARCHITECTURE:

Block Diagram of Deepfake Detection:

 


deepfake detection system can be broken down into several modules, each with its own specific functionality. Here is a high-level block diagram of the modules that may be included in a typical deepfakedetection system:

Data Collection: The first module in the deepfake detection system is data collection, where videos and images are collected from various sources, such as social media platforms, websites, and other online sources.

Pre-processing: The pre-processing module is responsible for preparing the collected data for further analysis. This may involve converting the data into a standard format, normalizing the data, and removing any noise or artifacts that may be present.

Feature Extraction: The feature extraction module is responsible for extracting meaningful features from the pre-processed data. This may include extracting facial landmarks, speech patterns, or other features that are unique to deepfakes.

Classification: The classification module is responsible for classifying the data as either real or deepfake. This is typically done using machine learning techniques such as supervised or unsupervised learning algorithms.

Verification: The verification module is responsible for verifying the authenticity of the classified data. This may involve checking the metadata associated with the data, such as the location, time, and device used to capture the data.

This algorithm will be used in deepfake detection.

a Convolutional Neural Network (CNN) algorithm and Long Short-Term Memory (LSTM) is used as an approach to detect the Deepfake videos.

Software model of Deepfake detection:


 

PROJECT MILESTONES AND DELIVERABLES:

Gantt PRO of Deepfake Detection Work:


List of milestones that could be included in a study.

Define the scope and objectives of the deepfakedetection project.

Gather and prepare a dataset of authentic and deepfakecontent.

Evaluate the performance of the baseline detection model and identify areas for improvement.

Develop and test techniques to detect deepfakes in real-time or near real-time, such as those that are distributed over live video feeds or social media platforms.

Develop and implement a user-friendly interface for the deepfake detection system.

Disclaimer: We will create deepfakes of BBSUL teacher/student with permission, also we will detect them content with the help of deepfake detection. 

 

 

Motivation OF Study:

The motivation of deepfake detection is primarily driven by the potential harms that deepfakes can cause. Deepfakes are AI-generated videos or images that can be used to manipulate public opinion, spread misinformation.

Protecting individuals and organizations: Deepfakescan be used to harm individuals and organizations by creating false representations of them or by spreading misinformation. Deepfake detection can help protect individuals and organizations from reputational damage or other harmful impacts.

Combating fake news and disinformation: Deepfakedetection can help prevent the spread of false information and enable more accurate reporting of news and events.

Improving media literacy: Deepfake detection can help improve media literacy by educating individuals on how to identify deepfakes and other forms of manipulated

 

 

REFERENCES:

• 1st Edition DeepFakes Creation, Detection, and Impact Edited By Loveleen Gaur Copyright Year 2023
• Deepfake Hardcover – October 6, 2020 by Sarah Darer Littman
• Deep Learning for Deepfakes Creation and Detection: A Survey (Research paper 2022)
• Computer Vision and Image Understanding Volume 223, October 2022, 103525
• Deepfake Detection through Deep Learning Publisher: IEEE Deng Pan; Lixian Sun; Rui Wang; Xingjian Zhang; Richard O. Sinnott

 

Comments

Popular posts from this blog

The AnyDesk Scandal: (18,000) Clients Information Was Disclosed And Listed For Sale On The Dark Web 2024

The Positive Power of Deepfake Technology

World's Largest Digital Camera Takes 3200 Megapixel Photos! (LSST) Large Synoptic Survey Telescope