DEEP LEARNING BASED

APPROACH FOR IMAGE

TAMPERING LOCALIZATION
IN IMAGE FORENSICS
Keerthi Prem R.
Reg no: TVE22ECAEI05
Guided by : Dr. Alex Raj S. M.
College of Engineering Trivandrum
Applied Electronics and Instrumentation
 AGENDA
Introduction
01
Selection of topic

Problem Statement
02
Reason for choosing this problem

03 Research Objective
Objective of the research

General Workflow
04
Workflow diagram

PRESENTATION TITLE
 Literature Survey
05
Literature review on this area

WorkPlan
06
Plan of implementation

Work Done so far..

07
Whats done so far

Conclusion
08
Conclusion of the project

References
09
Papers referred for the implementation of
the research.
PRESENTATION TITLE
Time Constraints

Introduction Working Work so far Conclusion

Appx 5 min Appx 18 min Appx 5 min Appx 5 min

 0
1
Introduction
 Introduction

• The availability of various image-editing software and tools

such as Photoshop, GIMP, etc., has made it possible to create
forgeries with minimal effort.

• Today, it is quite easy to produce a manipulated media that

looks indifferent to the human eyes. Various types of digital
image forgeries have evolved, the major ones include copy-
move, morphing, etc.

• But there is no reliable way of detecting and localize these

changes. We can create a model using Deep Learning to tackle
this problem.
 0
2
Research Objective
Research Objective

• To find effective and efficient techniques to accurately detect and identify

manipulated images.

• To detect various types of image forgeries such as copy-move, splicing,

and retouching.
 0
3
General Workflow
Workflow Diagram
 OUR
WORKFLOW

ELA Forgery
Preprocessing CNN Training/ Real/Fake? localization
Algorithm Testing
Image Forgery

Image
Image Forgery
Forgery

Dataset
Dataset ELA
ELA Algorithm
Algorithm Learning
Learning

Preprocessing Segmentation ELA analysis Comparison Testing Training/Validation

Output- Image
Fake/Real?? Tampering Localization
 Authentic Image after
performing ELA Algorithm
 Forged Image after
performing ELA Algorithm
Test Results of Tampering detection

Authentic Image tested and predicted as authentic

Test Results of Tampering detection
 0
5
Literature Survey
TITLE AUTHOR AND EXPLANATION
YEAR

IMAGE FORGERY Davide Cozzolino, The machine-learning is based on a

LOCALIZATION THROUGH Diego Gragnaniello, suitable local descriptor, while block-
THE FUSION OF CAMERA- Luisa Verdoliva matching relies on the PatchMatc
BASED, FEATURE-BASED algorithm. A decision fusion strategy is
AND PIXEL-BASED then imple_x0002_mented, based on
TECHNIQUES suitable reliability indexes associated
with the binary masks.

PRNU-Based Content XUFENG LIN AND Forgery localization based on photo-

Forgery Localization CHANG-TSUN LI response non-uniformity (PRNU)
Augmented With Image noise as an example and propose a
Segmentation segmentation-based forgery localization
scheme that exploits the local
homogeneity of visually indiscernible
clues to mitigate the limitations of
existing segmentation approaches that
are merely based on visually perceptible
content.
TITLE AUTHOR AND EXPLANATION
YEAR
SIMULTANEOUS FORGERY Paul Buchana*, By training the optimal trade-off
IDENTIFICATION AND Irina Cazan*, synthetic discriminant function
LOCALIZATION IN Manuel Diaz- (OTSDF) filter on each section of a
PAINTINGS USING Granados*, coarsely parceled image of an original
ADVANCED CORRELATION Felix Juefei-Xu painting, we are not only able to distinguish
FILTERS and Marios between a low_x0002_quality digitized
Savvides representation of a painting and its
forgery,but also specifically indicate where
the differences occur and where the replica
is particularly faithful to the original. This
method is also valuable in determining
whether an original painting has undergone
any modifications, given that a
representation of the initial version is
available.
Exposing Blurred Image Feng Zeng, Wei Novel method for blurred image splicing
Forgeries through Blind Wang*, Min localization based on blind image
Image Restoration Tang and restoration. We use blur parameters
Zhanghua Cao estimation through the spectrum
characteristics of blurred images in order to
restore the spliced region and the rest of the
TITLE AUTHOR AND EXPLANATION
YEAR
Copy-Move Forgery Yogitha, R. P, B. B. This research paper presents a novel
Localization Using DCT S, M. Reddy and R. approach to the above problem by using a
With LoG Filter Reddy LoG (Laplacian of Gaussian ) filter and
DCT (Discrete Cosine Transform) on the
image to detect a type of forgery known as
Copy-Move Forgery.
Image Forgery N. K. Hebbar and The forged images are processed with
Localization Using U-Net A. S. Kunte Error Level Analysis (ELA) which
based Architecture and highlights the forged region having
Error Level Analysis different compression levels and helps in
improving the model performance. The
processed images are used to train the U-
Net based encoder-decoder
architecture. A pre-trained Residual
Network 50 (ResNet50 backbone network
initialized with pre-trained weights is used
in the encoder of the network to speed up
the training process and enhance the
performance of the model by transfer
learning.The encoded features are
concatenated with the low-level image
 0
6
WorkPlan
WORK PLAN
Activity Estimated Date
Guide Assigned 3rd week of March 2023
1st week of November
Area Selection
2023
1st week of November
Topic Selection
2023
2nd week of November
1st Presentation
2023
1st week of December
WORK PLAN
Activity Estimated Date
3rd week of December
Localization Method
2023
3rd week of December
2nd Presentation
2023
4th week of December
Error Debugging
2023
1st week of January
DTFT implementation
2023
 0
7
Work Done so far..
 Work Done for Forgery detection

1. Conducted literature surveys and have tried to implement various

methods of Forgery detection like Segmentation, classification etc.
2. Finally concluded to proceed with ELA Algorithm.
3. Various dataset available for this was CASIO, SEAM,CMFD etc..
Downloaded dataset of CASIA 2 from Kaggle .Added few extra samples
to the dataset.
4. With the help of library functions like numpy,pandas,keras,matplotlib
and sklearn implemented with python in Anaconda.
5. With the help of ELA algorithm performed ELA analysis on Forged and
Un-Forged images.
6. Compared the output of both of them and algorithm when
implemented with CNN was able to perform accurately and correctly
classified them as forged and unforged.
 Work Done for Forgery Localization

1. Conducted literature surveys and have tried to implement various

methods of Forgery localization technique like PRNU, softlight, DCT
with Log filter etc..

2. Found out one of the effective localization to be implemented with

PIL library function in python.

3. Found a new method by taking the DFT of both ELA and Original image
and by understanding the coefficients and making superimposing
model to give accuracy.
 COMPARISON STUDY
Articl Type of Features, Model & # Data Set Pros/Cons Remarks
e No. Tampering Layers

1. Cut-paste, Copy- Three-level, 2-D CASIA v1.0, Advantage: 1. It will characterize Accuracy
move Daubechies wavelet CASIA v2.0, and tampered regions across JPEG and 91.09%
decomposition & SAE Columbia TIFF image formats. Disadvantages:
Stacked Autoencoders 1. The interfered areas must be
manually recognized. 2. The
interfered areas are not precisely
detected.
2. Cut-paste Image residuals & CASIA, COLUMB Advantage: 1. They attained the Accuracy 76%
RRU-Net tamper detection without any pre-
processing & post-processing
Disadvantages: 1. They have not
visualized the latent discriminative
feature between interfered and
untampered sections.
3 Image Splicing Surface probability & CASIA v1.0, Advantage: 1. The projected 0.61
edge probability map CASIA v2.0, approaches outperform present (Columbia)
& MFCN (Multi-task Columbia and splicing localization methods. MCC Score
fully convolution Carvalho Disadvantages: 1. They used the 0.52 (CASIA
network trained model to assess images, v1.0) & F1-
which are not in the training set. Score 0.54
(CASIA v1.0)
 COMPARISON STUDY
Articl Type of Features, Model & # Data Set Pros/Cons Remarks
e No. Tampering Layers

4 Combinations of CNN - pre-trained Dataset MICC- Advantage: 1. They exploited the Accuracy
geometrical and AlexNet Model F220 SVM as a classifier with the best 93.94%
transformations accuracy. Disadvantages: 1. Not
attacks to the more suitable for all data sets
original image

5 Median filtering Median filter residuals BOSS base 1.01, Advantage: 1. The outcomes show Accuracy
and Cut-paste & the CNN with 9 UCID, NRCS that the the suggested technique 85.14%
layers Photo Gallery, attains significant performance
Dresden, BOSS improvements. Disadvantages: 1. It
RAW (15352 is limited to recognize cutand-paste
images) imitations only.
6 Proposed Method CNN with ELA and CASIA v1.0, Advantage : 1, Identification of --
DTFT CASIA v2.0 compression levels.
2, Variation of threshold change in
detection to even minor changes.
Expected Output:
Obtained Output:
 0
8
Conclusion
 Conclusion

• In the proposed system, we are implemented different

concepts of image forgery detection with ELA algorithm and
CNN based method with CASIO 1&2 datasets.
• The system is capable of taking input image and give
out suitable outputs to solve the obstacle of forged
images.
• The system can be used in law and enforcements and cyber
security to help the user to differentiate between legitimate
and tampered images.
 0
9
References
 References
[1]N. Wang, L. Yi, G. Wang and X. Liu, "MemAU-Net: Memory-Enhanced Attention U-Net for
Medical Image Forgery Localization," 2023 International Joint Conference on Neural Networks
(IJCNN), Gold Coast, Australia, 2023, pp. 1-7, doi: 10.1109/IJCNN54540.2023.10191370.
[2]Yogitha, R. P, B. B. S, M. Reddy and R. Reddy, "Copy-Move Forgery Localization Using DCT
With LoG Filter," 2023 Third International Conference on Secure Cyber Computing and
Communication (ICSCCC), Jalandhar, India, 2023, pp. 658-664, doi:
10.1109/ICSCCC58608.2023.10176525.
[3]P. Buchana, I. Cazan, M. Diaz-Granados, F. Juefei-Xu and M. Savvides, "Simultaneous
forgery identification and localization in paintings using advanced correlation filters," 2016
IEEE International Conference on Image Processing (ICIP), Phoenix, AZ, USA, 2016, pp. 146-
150, doi: 10.1109/ICIP.2016.7532336.
[4]F. Zeng, W. Wang, M. Tang and Z. Cao, "Exposing Blurred Image Forgeries through Blind
Image Restoration," 2015 10th International Conference on P2P, Parallel, Grid, Cloud and
Internet Computing (3PGCIC), Krakow, Poland, 2015, pp. 466-469, doi:
10.1109/3PGCIC.2015.50.
[5]N. K. Hebbar and A. S. Kunte, "Image Forgery Localization Using U-Net based Architecture
and Error Level Analysis," 2021 3rd International Conference on Advances in Computing,
Communication Control and Networking (ICAC3N), Greater Noida, India, 2021, pp. 1992-
1996, doi: 10.1109/ICAC3N53548.2021.9725373.
 ReferencesAndrey. (2019). Digital image forgery detection using deep learning
[7][6]Kuznetsov,
approach. Journal of Physics: Conference Series. 1368. 032028. 10.1088/1742-
6596/1368/3/032028.
[8]Yogitha, R. P, B. B. S, M. Reddy and R. Reddy, "Copy-Move Forgery Localization Using
DCT With LoG Filter," 2023 Third International Conference on Secure Cyber Computing
and Communication (ICSCCC), Jalandhar, India, 2023, pp. 658-664, doi:
10.1109/ICSCCC58608.2023.10176525.
[9]Yang, Jiachen & Xiao, Shuai & Li, Aiyun & Lan, Guipeng & Wang, Huihui. (2021).
Detecting fake images by identifying potential texture difference. Future Generation
Computer Systems. 125. 10.1016/j.future.2021.06.043.
[10]Ahmad, Mohassin & Khurshid, Farida. (2022). Detection and localization of image
tampering in digital images with fused features. Concurrency and Computation:
Practice and Experience. 34. 10.1002/cpe.7191.
[11]S. Manjunatha. and M. M. Patil, "Deep learning-based Technique for Image Tamper
Detection," 2021 Third International Conference on Intelligent Communication
Technologies and Virtual Mobile Networks (ICICV), Tirunelveli, India, 2021, pp. 1278-
1285, doi: 10.1109/ICICV50876.2021.9388471.
THANK YOU

PRESENTATION TITLE