DEEPFAKE DETECTION

ABSTRACT
The image forgery is going on increasing day-by-day. In this project, we are performing a
review of this Image forgery technique. There are two kinds of techniques for image
forensics: one is active protection, and the other is passive detection. The main types of
Image forgery techniques are Image Splicing, Copy-Move forgery used mainly for making
tempered photographs are studied in more detail in this project. As the forgery of Images is
growing day-by-day, it is very much necessary to develop tools for detection as which image
is true and which is forgery. We study on of the most powerful technique like the SVM
classifier, Pixel-based and partition-based to detect forgery images

DEPARTEMENT OF INFORMATION SCIENCE & ENGINEERING

PDA COLLEGE OF ENGINEERING , KALABURAGI Page 1
 DEEPFAKE DETECTION

1. INTRODUCTION
As the use of images have been increasing day by day in our lives, with the introduction of
digital technology, The forgery of digital image has become more and more simple and
indiscoverable. Today's digital technology had begun to erode the integrity of images and
image counterfeiting and forgeries with the move to the world of Megapixels, opens a new
door to the dark-side of it. We are living in an age, where anything can be manipulated or
altered with the help of modern technology. With the increasing applications of digital
imaging, different types of software tools are introduced for processing images and
photographs. They are used to make forge images to make it look real or objects can be added
or deleted. For decades, photographs have been used to document and they have used as
evidence in courts. Although photographers are able to create composites of analog picture.
But this process is very time consuming and requires expert knowledge so it is hard to
implement than digital pictures. Today, however, powerful digital image editing software
makes image modifications straightforward [1]. Today’s digital technology has begun to
remove trust in our knowledge, as from the magazines, to fashion world and in scientific
journals, political campaigns, courts and the photo that come in our e-mail. In all of these
forged photographs are appearing with a more frequencies and sophistication. In the increase
in the availability of multimedia data in digital form has come to a tremendous growth of
tools to manipulate digital multimedia contents. The process of creating fake image has been
tremendously simple with the introduction of new and powerful computer graphics editing
software which are freely available as Photoshop, GIMP, and Corel Paint Shop. Today, this
powerful image processing software’s allow people to modify photos and images
conveniently and un perceivably. Now days it creates a big challenge to authenticate images.
Image forgery means manipulation of the digital image to conceal some meaningful or useful
information from it. Sometimes it is difficult to identify the edited region from the original
image. The detection of a forged image is driven by the need of authenticity and to maintain
integrity of the image. The survey has been done on existing techniques for forged image and
it highlights various copy–move detection and splicing detection methods based on their
robustness and computational complexity [2]. A forgery detection method that exploits
suitable inconsistencies in the color of the illumination of images. To achieve this, we
incorporate information from physics- and statistical-based illuminate estimators on image
regions. We try to extract texture- and edge-based features from the illuminate estimates.
These features are provided to a machine-learning approach for making decision

DEPARTEMENT OF INFORMATION SCIENCE & ENGINEERING

PDA COLLEGE OF ENGINEERING , KALABURAGI Page 2
 DEEPFAKE DETECTION

automatically. The classification performance using an SVM metafusion classifier is

promising. A SVM classifier is trained for using statistical features of pattern noise for
classifying smaller blocks of an image. SVM classifier is used which have similar functional
form to neural networks. Image, texture and pixel value based features are extracted and
analyzed from the images. Then has values are calculated for these features. The process
consists of two phases, which are training phase, and a testing phase. For each image in the
database the hash value is generated by image hashing. These hash values can be used for
content-based image retrieval, indexing image in database, and for authenticating, avoided
and mitigate the forgery of digital images [3], etc. To ensure trustworthiness, multimedia
authentication techniques have emerged to verify content integrity and prevent forgery.
Experimental results show that even slightest of image tempering can be detection with the
proposed technique can lead to provide authentication as the provided image is trusty.

DEPARTEMENT OF INFORMATION SCIENCE & ENGINEERING

PDA COLLEGE OF ENGINEERING , KALABURAGI Page 3
 DEEPFAKE DETECTION

2. LITERATURE SURVEY
1. Armas Vega, Esteban Alejandro, et al. Digital images authentication technique based on
DWT, 12authentication technique that combines the analysis of local texture patterns with
the discrete wavelet transform and the discrete cosine transform to extract features from
each of the blocks of an image. Subsequently, it uses a vector support machine to create a
model that allows verification of the authenticity of the image. Experiments were
performed with falsified images from public databases widely used in the literature that
demonstrate the efficiency of the proposed method
2. Vialatte, Jean-Charles, Vincent Gripon, and Gilles Coppin. Learning local receptive fields
and their weight sharing scheme on graphs. 2017 IEEE Global Conference on Signal and
Information Processing (GlobalSIP). IEEE, 2017, We propose a simple and generic layer
formulation that extends the properties of convolutional layers to any domain that can be
described by a graph. Namely, we use the support of its adjacency matrix to design
learnable weight sharing filters able to exploit the underlying structure of signals in the
same fashion as for images. The proposed formulation makes it possible to learn the
weights of the filter as well as a scheme that controls how they are shared across the
graph. We perform validation experiments with image datasets and show that these filters
offer performances comparable with convolutional ones.
3. Alhussein, Musaed. Image tampering detection based on local texture descriptor and
extreme learning machine. 2016 UKSim-AMSS 18th International Conference on
Computer Modelling and Simulation (UKSim). IEEE, 2016, This paper proposed a new
image tampering detection method based on local texture descriptor and extreme learning
machine (ELM). The image tampering includes both splicing and copy-move forgery.
First, the image was decomposed into three color channels (one luminance and two
Chroma), and each channel was divided into non-overlapping blocks. Local textures in
the form of local binary pattern (LBP) were extracted from each block. The histograms of
the patterns of all the blocks were concatenated to form a feature vector. The feature
vector was then fed to an ELM for classification. The ELM is a powerful and fast
classification approach. The experiments was performed using two publicly available
databases. The experimental results showed that the proposed method achieved a high
detection accuracy in both the databases.
4. Ojeniyi, Joseph A., et al. Hybridized Technique for Copy Move Forgery Detection Using
Discrete Cosine Transform and Speeded-Up Robust Feature Techniques. International

DEPARTEMENT OF INFORMATION SCIENCE & ENGINEERING

PDA COLLEGE OF ENGINEERING , KALABURAGI Page 4
 DEEPFAKE DETECTION

Journal of Image, Graphics and Signal Processing. 10(4)(2018), 22, As the world has
greatly experienced a serious advancement in the area of technological advancement over
the years, the availability of lots of sophisticated and powerful image editing tools has
been on the rise. These image editing tools have become easily available on the internet,
which has made people who are a novice in the field of image editing, to be capable of
tampering with an image easily without leaving any visible clue or trace behind, which
has led to increase in digital images losing authenticity. This has led to developing
various techniques for tackling authenticity and integrity of forged images. In this paper,
a robust and enhanced algorithm is been developed in detecting copy-move forgery,
which is done by hybridizing block-based DCT (Discrete Cosine Transform) technique
and a key point-based SURF (Speeded-Up Robust Feature)technique using the MATLAB
platform. The performance of the above technique has been compared with DCT and
SURF techniques as well as other hybridized techniques in terms of precision, recall, FPR
and accuracy metrics using MICC-F220 dataset. This technique works by applying DCT
to the forged image, with the main goal of enhancing the detection rate of such image and
then SURF is applied to the resulting image with the main goal of detecting those areas
that are been tampered with on the image. It has been observed that this paper’s technique
named HDS has an effective detection rate on the MICC-F220 dataset with multiple
cloning attacks and other various attacks such as rotation, scaling, a combination of
scaling plus rotation, blur, compression, and noise.
5. Ardizzone, Edoardo, Alessandro Bruno, and Giuseppe Mazzola. Copy–move forgery
detection by matching triangles of key points. IEEE Transactions on Information
Forensics and Security. 10(10)(2015), 2084-2094. — Copy–move forgery is one of the
most common types of tampering for digital images. Detection methods generally use
block-matching approaches, which first divide the image into overlapping blocks and then
extract and compare features to find similar ones, or point-based approaches, in which
relevant keypoints are extracted and matched to each other to find similar areas. In this
paper, we present a very novel hybrid approach, which compares triangles rather than
blocks, or single points. Interest points are extracted from the image, and objects are
modeled as a set of connected triangles built onto these points. Triangles are matched
according to their shapes (inner angles), their content (color information), and the local
feature vectors extracted onto the vertices of the triangles. Our methods are designed to
be robust to geometric transformations. Results are compared with a state-of-the-art block

DEPARTEMENT OF INFORMATION SCIENCE & ENGINEERING

PDA COLLEGE OF ENGINEERING , KALABURAGI Page 5
 DEEPFAKE DETECTION

matching method and a point-based method. Furthermore, our data set is available for use
by academic researchers.
6. Apollonio, F. I., et al. Evaluation of feature-based methods for automated network
orientation. International Archives of the Photogrammetry, Remote Sensing & Spatial
Information Sciences. 45(2014). Every day new tools and algorithms for automated
image processing and 3D reconstruction purposes become available, giving the possibility
to process large networks of unoriented and markerless images, delivering sparse 3D
point clouds at reasonable processing time. In this paper we evaluate some feature-based
methods used to automatically extract the tie points necessary for calibration and
orientation procedures, in order to better understand their performances for 3D
reconstruction purposes. The performed tests - based on the analysis of the SIFT
algorithm and its most used variants - processed some datasets and analysed various
interesting parameters and outcomes (e.g. number of oriented cameras, average rays per
3D points, average intersection angles per 3D points, theoretical precision of the
computed 3D object coordinates, etc.).

DEPARTEMENT OF INFORMATION SCIENCE & ENGINEERING

PDA COLLEGE OF ENGINEERING , KALABURAGI Page 6
 DEEPFAKE DETECTION

3. PROBLEM STATEMENT
The project focuses on applying widely used key point based algorithms in the field of object
recognition for CMFD. The proposed CMFD technique, consisting of both feature extraction
and feature matching, would serve as an alternative to the current state of the art CMFD
method using Speeded Up Robust Features (SURF). Before SURF was introduced in the field
of CMFD, Scale-invariant Feature Transform (SIFT) is widely regarded as one of the best
key point based algorithm for CMFD. With SURF, the algorithm proposed multiple
optimization which successfully reduced the computation time by 3 times. However, the
improvement came at the expense of the accuracy rate. Hence, this work aims to identify a
Copy Move Forgery Detection technique, using machine learning.

DEPARTEMENT OF INFORMATION SCIENCE & ENGINEERING

PDA COLLEGE OF ENGINEERING , KALABURAGI Page 7
 DEEPFAKE DETECTION

4. EXISTING SYSTEM
This existing system tries to present the image copy detection model for the 2 most common
image tampering method which includes copy-image forgery detection and Image Splicing.
Also Feature extraction process is done through the Speeded up robust features(SURF), and
Scale Invariant Feature Transform(SIFT) Descriptor method for identifying the matching
point.
In the case of Image Splicing Detection, it is used for extracting the image edges of
Y(brightness in luma), Cb(blue minus luma), and Cr (Red minus luma) of the image
components.
Gray Level Co-Occurrence Matrix (GLCM) is utilized for each and every edge internal
image through which the feature vector is created. The created feature vector is fed into a
Support Vector Machine (SVM Classifier).
In the existing system, the method for image copy detection represents an outcome showing
that the Speeded up robust feature is superior to the Scale Invariant Feature Transform.
The method simulates 80 percent accuracy for the detection of tampered images. The image
processing technique using the YCb Cr color model showed a significant result in image
splicing detection. The outcome from this method simulates 98 percent positivity for the
detection of image splicing.

DEPARTEMENT OF INFORMATION SCIENCE & ENGINEERING

PDA COLLEGE OF ENGINEERING , KALABURAGI Page 8
 DEEPFAKE DETECTION

5. DISADVANTAGES OF EXISTING SYSTEM

1. Less accuracy:Less accuracy refers to a situation in which a model or system fails to
correctly predict or classify outcomes in comparison to a desired or expected level of
precision. Less accuracy could result from various factors, including issues with the
quality or quantity of training data, inappropriate model architecture, insufficient model
training, or challenges in the nature of the problem being addressed.
2. Constantly Evolving Techniques: Deepfake creation techniques are rapidly evolving, and
new methods are regularly introduced. Detection systems may struggle to keep pace with
these advancements, leading to a continuous cat-and-mouse game between creators and
detectors.
3. Generalization Issues: Deepfake detectors may be trained on specific datasets, making
them effective against certain types of deepfakes but less reliable against new or unseen
variations. Generalizing across different types of deepfakes is a significant challenge.
4. Resource Intensive: Some deepfake detection methods, especially those based on
complex machine learning models, can be computationally intensive. This may limit their
real-time applicability, especially in situations where rapid and efficient processing is
crucial.
5. Limited Training Data: Developing robust deepfake detection models requires a diverse
and extensive dataset. Obtaining a comprehensive dataset that covers all possible
variations and scenarios is challenging, potentially limiting the effectiveness of detection
systems.
6. Real-time Constraints: Some applications, such as live video streaming, demand real-
time deepfake detection. Achieving low-latency detection without compromising
accuracy is a significant challenge.

DEPARTEMENT OF INFORMATION SCIENCE & ENGINEERING

PDA COLLEGE OF ENGINEERING , KALABURAGI Page 9
 DEEPFAKE DETECTION

6. OBJECTIVES

 To identify a Copy Move Forgery Detection technique like the SVM classifier, Pixel-based
and partition-based to detect forgery images
 To find whether a given image or video is real or fake
 To find signature is fake or not
 Object Recognition: Using Haar cascade and SVM for object recognition in images or videos
can be beneficial in applications like surveillance, robotics, or autonomous vehicles.
 Facial Recognition: SVM classifiers can be trained to recognize faces, which has applications
in security, access control systems, and human-computer interaction.
 Quality Control in Manufacturing: SVM classifiers can assist in quality control by identifying
defects or anomalies in manufacturing processes.

DEPARTEMENT OF INFORMATION SCIENCE & ENGINEERING

PDA COLLEGE OF ENGINEERING , KALABURAGI Page 10
 DEEPFAKE DETECTION

7. PROPOSED SYSTEM
In this proposed system, giving the classification of the Image forgery approaches, which
mainly consist of Active approach and Passive approach. Introducing three types of image
forgery techniques. Providing the possible image tempering detection techniques based on the
SVM classifier. Haar cascade is used for detection of face parts. In preprocessing to remove noise
purpose fast-n denoised colored algorithm used. For extracting the features purpose Canny-edge
algorithm is used. Additionally with detection of fake or real human face also finding signature
is fake or real.

DEPARTEMENT OF INFORMATION SCIENCE & ENGINEERING

PDA COLLEGE OF ENGINEERING , KALABURAGI Page 11
 DEEPFAKE DETECTION

8. ADVANTAGES
1. Real-Time Processing: Haar cascade is known for its efficiency, making it suitable for real-
time object detection applications.
2. Pre-trained Models: Pre-trained Haar cascade models are available for common objects,
enabling quick implementation without the need for extensive training datasets.
3. Feature Extraction: Canny edge detection is effective in extracting prominent features from
images, which can be valuable for subsequent analysis or processing.
4. Edge Preservation: Canny algorithm tends to preserve important edges in an image, which
can be beneficial in applications like image segmentation
5. Binary Classification: SVMs are particularly effective for binary classification tasks, where
the goal is to separate data into two classes.
6. Robust to Over fitting: SVMs are less prone to overfitting, making them suitable for
scenarios with limited training data

DEPARTEMENT OF INFORMATION SCIENCE & ENGINEERING

PDA COLLEGE OF ENGINEERING , KALABURAGI Page 12
 DEEPFAKE DETECTION

9. SYSTEM REQUIREMENTS

Software Requirement
Operating System : Windows 7/10 or above
Front End : PYTHON
Back End : SQLITE3

Hardware Requirement
Processor : Intel Core I5 and above
Processor Speed : 1.0GHZ or above
RAM : 4 GB RAM or above
Hard Disk : 500 GB hard disk or above

DEPARTEMENT OF INFORMATION SCIENCE & ENGINEERING

PDA COLLEGE OF ENGINEERING , KALABURAGI Page 13
 DEEPFAKE DETECTION

10. SYSTEM DESIGN

Figure: architecture of deepfake using svm classifier

The image forgery detection system is usually a binary decision problem i.e., identifying
whether an image is authentic or forged. The system requires a pre-processing step to extract
features that are followed by a classification step. The statistical and structural changes
because of tampering are reflected by the features of an image. Therefore, an effective feature
extraction technique is highly important, which is the main focus of our proposed system.
The extracted features are fed into a binary classifier (e.g., machine learning model) to
identify authentic and forged images.

DEPARTEMENT OF INFORMATION SCIENCE & ENGINEERING

PDA COLLEGE OF ENGINEERING , KALABURAGI Page 14
 DEEPFAKE DETECTION

Process of image recognisation

(b) Input Image (b) Gray Image

(c) Denoise Image

DEPARTEMENT OF INFORMATION SCIENCE & ENGINEERING

PDA COLLEGE OF ENGINEERING , KALABURAGI Page 15
 DEEPFAKE DETECTION

Figure: An Example Of An Adversarial Attack on A Deepfake Detect

DEPARTEMENT OF INFORMATION SCIENCE & ENGINEERING

PDA COLLEGE OF ENGINEERING , KALABURAGI Page 16
 DEEPFAKE DETECTION

11. CONCLUSION
In this project, we studied that, due to the advancement in the digital software’s manipulation
of digital images has become easy. As powerful computers, advanced photo-editing software
packages and high resolution capturing devices are invented. Out of all the cases of digital
image forgery, they can be categorized into two major groups as active and passive
approaches, based on the process involved in creating the fake image. We studied the types of
image forgery as, Image Retouching, Image Splicing, and Copy-Move Attack. As it is the
vital need to make trust.

While Haar cascade, Canny algorithm, and SVM classifier provide a foundation for
computer vision-based object detection and classification, their application in deepfake
detection has inherent limitations. Deepfake detection is a challenging task that often requires
more advanced techniques, such as deep learning approaches involving convolutional neural
networks (CNNs) and recurrent neural networks (RNNs). The dynamic and evolving nature
of deepfake creation necessitates continuous research and development of more sophisticated
and adaptable detection methods to stay ahead of deceptive practices. Integrating traditional
methods with advanced deep learning techniques could offer a more comprehensive and
robust solution for addressing the challenges posed by deepfake content.

DEPARTEMENT OF INFORMATION SCIENCE & ENGINEERING

PDA COLLEGE OF ENGINEERING , KALABURAGI Page 17
 DEEPFAKE DETECTION

REFERENCES
1. Sudha, N., and PravinKakar. Exposing postprocessed copy–paste forgeries through
transform-invariant features.
2. William, Youssef, SherineSafwat, and Mohammed AM. Salem. Robust Image Forgery Detection
Using Point Feature Analysis. 2019 Federated Conference on Computer Science and Information
Systems
3. Muhammad, Ghulam, et al. Image forgery detection using steerable pyramid transform and local
binary pattern. Machine Vision and Applications.
4. Kuznetsov, Andrey, and Vladislav Myasnikov. A new copy-move forgery detection algorithm
using image preprocessing procedure. Procedia engineering.
5. Bay, Herbert, et al. Speeded-up robust features (SURF). Computer vision and image
understanding.
6. Kanagavalli, N., and L. Latha. A survey of copy-move image forgery detection techniques. 2017
International Conference on Inventive Systems and Control (ICISC).

DEPARTEMENT OF INFORMATION SCIENCE & ENGINEERING

PDA COLLEGE OF ENGINEERING , KALABURAGI Page 18