Deep Fake Detection using

Deep Learning
 ABSTRACT

 This study focuses on addressing the rising challenges of deep fake detection, a critical issue in the digital age due to the proliferation of
manipulated media such as videos and images. The project leverages deep learning algorithms to identify deep fakes, which pose significant
risks in contexts like fake news, celebrity impersonations, and scams.

 A comprehensive evaluation was performed using a Kaggle deep fake dataset, testing multiple deep learning models including
InceptionResnetV2, VGG19, CNN, Xception, NasNetMobile, and an ensemble approach combining Xception and NasNetMobile. The proposed
ensemble model demonstrated exceptional performance, achieving an accuracy of 99.9955%, along with optimal precision, recall, and F1-score.

 This research highlights the potential of advanced deep learning models, particularly the ensemble approach, in providing a robust solution for
detecting deep fakes, offering a promising path toward mitigating the risks posed by manipulated digital content. The results underscore the
importance of continuous advancements in detection mechanisms to safeguard against the harmful impact of deep fakes in various domains.
 INTRODUCTION
 In the realm of technological evolution, the human face stands as the quintessential identifier, a uniquely personal aspect that defines
individuals. However, the relentless progress of face synthesis technologies has ushered in an era where the sanctity of facial identity is
under siege. The escalating threat of unauthorized face alterations looms large, prominently exemplified by the advent of deep fake
technology. This artificial intelligence marvel facilitates the seamless superimposition of one person's face onto another's, all executed
without the subject's consent. As a consequence, the profound implications of deep fakes extend far beyond the realm of mere privacy
invasion, posing a significant challenge to the integrity of personal and public security.

 At the forefront of combating this emerging menace is the application of deep learning—a formidable and versatile technology that has
left an indelible mark across various domains, including machine learning, computer vision, and natural language processing. With its
ability to discern intricate patterns and nuances, deep learning has emerged as a potent tool in the ongoing battle against deep fake
proliferation. This transformative technology holds the promise of fortifying our defenses against malicious face alterations, exemplifying
the innovative synergy between artificial intelligence and the safeguarding of our fundamental right to identity and security.
 OBJECTIVE

 The objective of this study is to develop a robust deep fake detection system using advanced deep
learning algorithms.

 By evaluating various models, including InceptionResnetV2, VGG19, CNN, Xception, NasNetMobile,

 The study seeks to propose a reliable detection mechanism that can be applied to safeguard against the
harmful implications of deep fakes in digital content.
 PROBLEM STATEMENT

 The rise of deep fake technology is driven by advancements in AI and deep learning, allowing for the creation of highly
convincing manipulated media, leading to the spread of misinformation and deception.

 Deep fakes, which involve altering videos and images to impersonate real individuals, have become increasingly sophisticated,
posing a significant challenge to media authenticity and digital security in various industries.

 Celebrities, politicians, and public figures are particularly vulnerable, as deep fakes can damage their reputation, spread fake news,
or even be used for blackmail, affecting public trust and personal privacy.

 The widespread use of deep fakes undermines the credibility of online content, fuels social manipulation, increases
misinformation, and may harm individuals or organizations by tarnishing reputations and inciting public unrest.

 This project aims to develop a deep learning-based system that leverages advanced algorithms for accurate detection of deep fakes,
helping identify manipulated media and mitigate the risks associated with this growing issue.
 LITERATURE SURVEY

 WildDeepfake: A Challenging Real-World Dataset for Deepfake Detection:

 WildDeepfake: A Challenging Real-World Dataset for Deepfake Detection (arxiv.org)

 ABSTRACT: In recent years, the abuse of a face swap technique called deepfake [2] has raised enormous public concerns. So far, a large
number of deepfake videos (known as "deepfakes") have been crafted and uploaded to the internet, calling for effective countermeasures. One
promising countermeasure against deepfakes is deepfake detection. Several deepfake datasets have been released to support the training and
testing of deepfake detectors, such as DeepfakeDetection [1] and FaceForensics++ [23]. While this has greatly advanced deepfake detection,
most of the real videos in these datasets are filmed with a few volunteer actors in limited scenes, and the fake videos are crafted by researchers
using a few popular deepfake softwares. Detectors developed on these datasets may become less effective against real-world deepfakes on the
internet. To better support detection against real-world deepfakes, in this paper, we introduce a new dataset WildDeepfake, which consists of
7,314 face sequences extracted from 707 deepfake videos collected completely from the internet.
 LITERATURE SURVEY

 Learning Self-Consistency for Deepfake Detection:

 Learning Self-Consistency for Deepfake Detection | IEEE Conference Publication | IEEE Xplore

 ABSTRACT: We propose a new method to detect deepfake images using the cue of the source feature inconsistency
within the forged images. It is based on the hypothesis that images’ distinct source features can be preserved and
extracted after going through state-of-the-art deepfake generation processes. We introduce a novel representation
learning approach, called pair-wise self-consistency learning (PCL), for training ConvNets to extract these source
features and detect deepfake images. It is accompanied by a new image synthesis approach, called inconsistency image
genera-tor (I2G), to provide richly annotated training data for PCL. Experimental results on seven popular datasets
show that our models improve averaged AUC over the state of the art from 96.45% to 98.05% in the in-dataset
evaluation and from 86.03% to 92.18% in the cross-dataset evaluation.
 LITERATURE SURVEY

 Detecting Deepfake Videos using Attribution-Based Confidence Metric:

 Detecting Deepfake Videos using Attribution-Based Confidence Metric | IEEE Conference Publication | IEEE Xplore

 ABSTRACT: Recent advances in generative adversarial networks have made detecting fake videos a challenging task. In this paper, we
propose the application of the state-of-the-art attribution based confidence (ABC) metric for detecting deepfake videos. The ABC
metric does not require access to the training data or training the calibration model on the validation data. The ABC metric can be used
to draw inferences even when only the trained model is available. Here, we utilize the ABC metric to characterize whether a video is
original or fake. The deep learning model is trained only on original videos. The ABC metric uses the trained model to generate
confidence values. For, original videos, the confidence values are greater than 0.94.
 LITERATURE SURVEY

 Adversarial Perturbations Fool Deepfake Detectors:

 Adversarial Perturbations Fool Deepfake Detectors | IEEE Conference Publication | IEEE Xplore

 ABSTRACT: This work uses adversarial perturbations to enhance deepfake images and fool common deepfake detectors. We created
adversarial perturbations using the Fast Gradient Sign Method and the Carlini and Wagner L 2 norm attack in both blackbox and
whitebox settings. Detectors achieved over 95% accuracy on unperturbed deepfakes, but less than 27% accuracy on perturbed
deepfakes. We also explore two improvements to deep-fake detectors: (i) Lipschitz regularization, and (ii) Deep Image Prior (DIP).
Lipschitz regularization constrains the gradient of the detector with respect to the input in order to increase robustness to input
perturbations. The DIP defense removes perturbations using generative convolutional neural networks in an unsupervised manner.
Regularization improved the detection of perturbed deepfakes on average, including a 10% accuracy boost in the blackbox case. The
DIP defense achieved 95% accuracy on perturbed deepfakes that fooled the original detector while retaining 98% accuracy in other
cases on a 100 image subsample.
 LITERATURE SURVEY

 DeepfakeStack: A Deep Ensemble-based Learning Technique for Deepfake Detection:

 DeepfakeStack: A Deep Ensemble-based Learning Technique for Deepfake Detection | IEEE Conference Publication | IEEE Xplore

 ABSTRACT: Recent advances in technology have made the deep learning (DL) models available for use in a wide variety of novel
applications; for example, generative adversarial network (GAN) models are capable of producing hyper-realistic images, speech, and
even videos, such as the so-called “Deepfake” produced by GANs with manipulated audio and/or video clips, which are so realistic as
to be indistinguishable from the real ones in human perception. Aside from innovative and legitimate applications, there are numerous
nefarious or unlawful ways to use such counterfeit contents in propaganda, political campaigns, cybercrimes, extortion, etc. To meet the
challenges posed by Deepfake multimedia, we propose a deep ensemble learning technique called DeepfakeStack for detecting such
manipulated videos. The proposed technique combines a series of DL based state-of-art classification models and creates an improved
composite classifier. Based on our experiments, it is shown that DeepfakeStack outperforms other classifiers by achieving an accuracy
of 99.65% and AUROC of 1.0 score in detecting Deepfake. Therefore, our method provides a solid basis for building a Realtime
Deepfake detector.
 EXISTING SYSTEM

 The existing system focuses on the analysis and utilization of previously deployed deep fake detection algorithms. It
extensively explores classic detection methods and contemporary deep learning-based approaches, including
Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), and Long Short-Term Memory
(LSTM).

 Classic methods often rely on rule-based heuristics, while deep learning methods harness the hierarchical learning
capabilities of neural networks to discern intricate patterns indicative of deep fakes. The study aims to provide a
comprehensive understanding of the strengths and limitations of these methods in identifying manipulated content.

 In contrast, the proposed system takes a proactive stance by integrating cutting-edge deep learning algorithms. The
proposed system aims to enhance detection accuracy, contributing to a deeper comprehension of deep fake
production and distribution for more effective countermeasures.
 DISADVANTAGES FOR EXISTING
SYSTEM

 The existing system primarily relies on classic rule-based heuristics, which often fail to capture complex patterns and
may have limited accuracy in detecting advanced deep fakes created using sophisticated deep learning techniques.

 While it employs basic deep learning algorithms like CNN, RNN, and LSTM, these methods lack the flexibility and
robustness needed to handle increasingly intricate deep fake manipulations, leading to potential limitations in detection.

 The existing system does not leverage ensemble models, which combine multiple algorithms to improve accuracy and
robustness in deep fake detection, potentially resulting in lower detection performance compared to more advanced
methods.

 The focus on classic detection methods may make the existing system less adaptable to emerging trends in deep fake
production, requiring frequent updates and refinements to keep pace with evolving techniques in media manipulation.
 PROPOSED SYSTEM

 The proposed system aims to develop a highly effective deep fake detection framework using a combination of advanced deep
learning models. It will incorporate multiple algorithms, including InceptionResnetV2, VGG19, CNN, Xception, and
NasNetMobile, to analyze and identify manipulated images and videos.

 Additionally, an ensemble approach, combining Xception and NasNetMobile, will be explored to enhance detection accuracy.
The system will be trained and evaluated using a publicly available Kaggle deep fake dataset, which includes a wide variety of
real and manipulated media samples.

 By leveraging the strengths of different architectures, the system seeks to optimize the performance of deep fake detection,
addressing the challenge of distinguishing between genuine and falsified content. The proposed approach will not only improve
the detection of deep fakes but also provide a scalable, reliable solution for real-world applications in media, cybersecurity, and
digital content verification, thereby reducing the harmful effects of deep fakes on individuals and society.
 ADVANTAGES FOR PROPOSED SYSTEM

 The proposed system integrates advanced deep learning models like InceptionResnetV2, VGG19, Xception, and
NasNetMobile, significantly enhancing detection accuracy and providing a more robust solution for identifying
complex deep fake content.

 By incorporating an ensemble approach that combines Xception and NasNetMobile, the proposed system optimizes
detection performance, leveraging the strengths of both algorithms to improve generalization and accuracy in
recognizing manipulated media.

 The system uses a publicly available Kaggle deep fake dataset for training and evaluation, ensuring a diverse range of
real and manipulated samples for better model generalization and practical applicability in real-world scenarios.

 The proposed system is scalable and reliable, making it suitable for deployment in various applications, including
media, cybersecurity, and digital content verification, thereby reducing the potential societal harm caused by deep fake
content.
 REQUIREMENTS

SOFTWARE HARDWARE
 Software : Anaconda  1) Operating System : Windows Only
 Primary Language : Python
 Frontend Framework : Flask  2) Processor : i5 and above
 Back-end Framework : Jupyter Notebook
 3) Ram : 8gb and above
 Database : Sqlite3
 Front-End Technologies :  4) Hard Disk : 25 GB in local drive
HTML,CSS,JavaScript and Bootstrap4
SYSTEM AECHITECTURE
UML DIAGRAMS
DATAFLOW DIAGRAM
USECASE DIAGRAM
CLASS DIAGRAM
ACTIVITY DIAGRAM
SEQUENCE DIAGRAM
COLLABORATION DIAGRAM
COMPONENT DIAGRAM
DEPLOYMENT DIAGRAM

User System
 MODULES

 Data exploration: using this module we will load data into system

 Image processing: Using the module we will process of transforming an image into a digital form and
performing certain operations to get some useful information from it.

 Splitting data into train & test: using this module data will be divided into train & test

 Model generation: Model building - InceptionResnetV2 - VGG19 - CNN – Xception – NasNetMobile –

 User signup & login: Using this module will get registration and login

 User input: Using this module will give input for prediction

 Prediction: final predicted displayed

 InceptionResnetV2 is a convolutional neural network architecture that combines the strengths of Inception modules and residual connections to improve the efficiency

of deep learning models. It utilizes a hybrid structure that incorporates both deep residual learning and multi-level feature extraction, allowing it to effectively capture

complex patterns in visual data. Its purpose is to enhance classification performance, especially for tasks involving large, high-dimensional datasets, by enabling faster

convergence and more accurate predictions in image recognition and manipulation detection tasks.

 VGG19 is a deep convolutional neural network architecture known for its simplicity and effectiveness in image classification tasks. It consists of 19 layers, including

convolutional layers, pooling layers, and fully connected layers, making it capable of extracting hierarchical features from input images. Its usage focuses on high-level

image classification tasks, where the network is trained to recognize various objects and patterns. The purpose of VGG19 is to offer a highly interpretable and scalable

model for feature extraction in visual recognition tasks.

 A Convolutional Neural Network (CNN) is a deep learning model designed for processing structured grid data, particularly images. It utilizes convolutional layers to

automatically learn spatial hierarchies of features from the input data, making it ideal for visual recognition tasks. CNNs are widely used in object detection, image

segmentation, and video analysis. The purpose of CNNs is to detect and classify patterns in visual data by applying convolution operations followed by pooling and fully

connected layers for final classification.

 Xception is an advanced convolutional neural network architecture that extends the Inception model by introducing depthwise separable convolutions. This structure
improves efficiency by reducing the computational complexity without sacrificing model performance. Xception's design is based on the hypothesis that depthwise
separable convolutions can achieve higher accuracy and faster training compared to traditional convolutions. It is primarily used for image classification, object
detection, and video analysis, offering enhanced performance by focusing on more effective feature extraction and parameter optimization.

 NasNetMobile is a mobile-optimized deep learning architecture developed using neural architecture search (NAS) to identify the most efficient network configuration
for mobile devices. It is designed to balance model performance and computational efficiency, making it ideal for resource-constrained environments. The purpose of
NasNetMobile is to provide accurate image classification while minimizing the model’s size and computational requirements, ensuring fast inference times. Its usage
spans mobile applications and embedded systems where efficient, real-time deep learning predictions are necessary, such as in visual recognition tasks.

 The Ensemble model combining Xception and NasNetMobile leverages the strengths of both architectures to improve prediction accuracy and robustness. By
combining the output of these models, the ensemble approach reduces the chances of overfitting and enhances generalization across different datasets. This method is
used for tasks requiring high accuracy, such as deep fake detection, by exploiting the complementary features learned by both models. The purpose of the ensemble is
to capitalize on the diversity of Xception’s feature extraction and NasNetMobile’s mobile efficiency, achieving optimal performance.
 CONCLUSION

 In conclusion, the deep fake detection system developed in this study demonstrates significant promise in identifying
manipulated media using advanced deep learning algorithms.

 Among the various models tested, the ensemble approach combining Xception and NasNetMobile emerged as the
highest-performing algorithm, achieving a remarkable accuracy of 99.9955%, with corresponding precision, recall, and
F1-score values that further highlight its efficacy.

 The ensemble method's superior performance can be attributed to its ability to harness the strengths of both models, resulting
in enhanced generalization and robust detection capabilities. This high accuracy suggests that the proposed system is well-
suited to address the growing concerns surrounding deep fake content and can be deployed in real-world scenarios to
safeguard against the risks posed by digital manipulations.

 By focusing on optimizing detection methods, the research offers a practical solution to mitigate the harmful impact of deep
fakes in various sectors, including media, politics, and cybersecurity.
 FUTURE SCOPE

 Future work will focus on enhancing the system's ability to handle more diverse and complex datasets,
including videos with varying resolutions, lighting conditions, and dynamic scenes.

 This will involve fine-tuning the model architectures and exploring additional data augmentation
techniques to improve generalization. Furthermore, the integration of multi-modal detection (audio, text,
and visual analysis) could be explored to detect deep fakes in a broader range of contexts.

 Continuous evaluation and adaptation to emerging deep fake techniques will ensure the system's
robustness and relevance in real-world applications.
 REFERENCES

 [1] M. Mirza and S. Osindero, “Conditional generative adversarial nets,” arXiv preprint arXiv:1411.1784, 2014.

 [2] Y. Bengio, P. Simard, and P. Frasconi, “Long short-term memory,” IEEE Trans. Neural Netw, vol. 5, pp. 157–166, 1994.

 [3] I. Goodfellow, Y. Bengio, and A. Courville, Deep learning. MIT press, 2016.

 [4] S. Hochreiter, “Ja1 4 rgen schmidhuber (1997).“long short-term memory”,” Neural Computation, vol. 9, no. 8.

 [5] M. Schuster and K. Paliwal, “Networks bidirectional reccurent neural,” IEEE Trans Signal Proces, vol. 45, pp. 2673–2681,

1997.