This document discusses techniques for detecting digital image forgeries. It begins by defining different types of forgeries such as image retouching, splicing, and cloning. It then discusses mechanisms for forgery detection, distinguishing between active methods that embed hidden information in images and passive methods that analyze image traces. A key technique presented is using rotation angle estimation to detect cloned regions, with details on calculating variance to determine the rotation angle. The document concludes by presenting an algorithm for region duplication detection using hybrid wavelet transforms like DCT, Walsh, and Hadamard transforms.

1. 1

2. Digital Image Forgery
Detection
2

3. Presentation Division
Introduction
Forgery
Detection
Region
Duplication
Conclusion
 Digital Image
Forgery
Detection
 Types of
Forgery
 Forgery
Detection
Mechanisms
 High Precision
Rotation Angle
Estimation For
Copy Move.
 Explaination
 Rotation Angle
Calculation
 Variance
Estimation
 Algorithm
 Discrete
Cosine
Transform
 Walsh
Transform
 Hybrid
Wavelet
Transform
 Results
 Future Works
 References
3

4. Digital Image Forgery Detection
 Alteration of the semantic components of a digital
image.
 Removing Contents from the image
 Adding Data to the image
 Types of Forgery
 Image Retouching
 Image Splicing (Copy-Paste)
 Image Cloning (Copy-Move)
4

5. Image Retouching
 One of the oldest types of image forgery
 Image features are tampered with.
 Used to enhance or reduce digital image features.
 Considered less dangerous type of image forgery.
5

6. Image Splicing (Copy-Paste)
 Fragments of 2 or more images are combined to form an image.
 This operation is fundamental in digital photo montaging and in turn is a
mechanism for image forgery creation.
 Image splicing technique may change the visual message of digital images
more aggressively than image retouching.
6

7. Image Cloning (Copy-Move)
 Considered as a special case of image splicing, where the tampering occurs
within a single image and no need for multiple images.
 Part of the image is copied and then pasted in a desired location within the
same image.
 The purpose of such tampering is to duplicate or conceal a certain object in
that image.
7

8. Image Cloning
 Blurring is usually used to reduce the expected irregularity along the border
of the pasted regions.
 The similarity of texture, color, noise and other information inside the image
make it very difficult to detect this kind of tampering via visual inspection.
 Moreover, performing of post-processing operations such as blurring, adding
noise and JPEG compression or geometric operations such as scaling, shifting
and rotation increase the hardness of detection task.
8

9. Forgery Detection Mechanisms
 Can be Classified into Two Types
 Active Methods
 Passive Methods
 Active Methods
 Hidden Information inside the Digital Image.
 Done at the time of Data Acquisition or before disseminated
to the public.
 Embedded information can be used to identify the source of
such image or to detect possible modification to that image.
9

10. Forgery Detection Mechanisms
(Active Methods)
 Two Major Types
 Digital Signature
 Digital Watermarking
10

11. Forgery Detection Mechanisms
(Passive Methods)
 Use traces left by the processing steps in different phases of acquisition and
storage of digital images.
 These traces can be treated as a fingerprint of the image source device.
 Passive methods work in the absence of protecting techniques.
 They do not use any pre-image distribution information inserted into digital
image.
 They work by analyzing the binary information of digital image in order to
detect forgery traces, if any
 Limitation is the number of false positives.
11

12. High Precision Rotation Angle Estimation for
Rotated Images
 Paper addresses the detection of “copy-move”(cloning) technique
 As discussed before cloning detection becomes harder when the forger uses
geometric alterations like scaling, rotation & shifting.
 Particularly addresses the Rotation transformation.
 This paper proposes a novel blind image rotation detection algorithm with
high precision rotation angle estimation
12

13.  S, t=pixel coordinates in the rotated image I.
 𝜑 = weighted value.
𝑖′
= 𝑖𝑐𝑜𝑠 𝜃 − 𝑗𝑠𝑖𝑛 𝜃
𝑗′
= 𝑖𝑠𝑖𝑛𝜃 + 𝑗𝑐𝑜𝑠 𝜃
I= Original image
I’=Intermediate Image
I”= Rotated Image
𝐼𝑠,𝑡
"
=
𝑛=−𝑁
𝑁
𝑚=−𝑁
𝑁
𝜑(𝑖 𝑠,𝑡
′
+ 𝑛, 𝑗𝑠,𝑡
′
+ 𝑚 )𝑰′(𝑖 𝑠,𝑡
′
+ 𝑛, 𝑗𝑠,𝑡
′
+ 𝑚
High Precision Rotation Angle Estimation for
Rotated Images
13

14.  α = horizontal distance rotated image I′′ & intermediate image I′
 β = vertical distance.
𝛼2
= (pcosθ − 𝑓𝑙𝑜𝑜𝑟(𝑝𝑐𝑜𝑠𝜃) + 𝑅)2
𝛽2
= (𝑞𝑠𝑖𝑛θ − 𝑓𝑙𝑜𝑜𝑟(𝑞𝑠𝑖𝑛𝜃) + 𝑆)2
R and S are constant(translation)
High Precision Rotation Angle Estimation for
Rotated Images
14

15. 𝑉𝑎𝑟 𝑋 = 𝐸[ 𝑋 − 𝜇 2
]
𝑉𝑎𝑟 𝑎𝑋 + 𝑏𝑌 = 𝑎2 𝑉𝑎𝑟 𝑋 + 𝑏2 𝑉𝑎𝑟(𝑌)
𝐼𝑠,𝑡
"
=
𝑛=−𝑁
𝑁
𝑚=−𝑁
𝑁
𝜑(𝑖 𝑠,𝑡
′
+ 𝑛, 𝑗𝑠,𝑡
′
+ 𝑚 )𝑰′(𝑖 𝑠,𝑡
′
+ 𝑛, 𝑗𝑠,𝑡
′
+ 𝑚
For a single pixel, we have:
𝐼𝑠,𝑡
"
≈
𝑛=0
1
𝑚=0
1
𝜑(𝑖 𝑠,𝑡
′
+ 𝑛, 𝑗𝑠,𝑡
′
+ 𝑚 )𝑰′(𝑖 𝑠,𝑡
′
+ 𝑛, 𝑗𝑠,𝑡
′
+ 𝑚
High Precision Rotation Angle Estimation for
Rotated Images
15

16.  Plot of horizontal distance vector and
its spectrum at 𝜃 = 300.
 Plot of peak frequency of distance vector
𝛼2
against all 𝜃 𝜖 [00
− 450
]. Frequency is
normalized to [0,1].
High Precision Rotation Angle Estimation for
Rotated Images
16

17.  Algorithm of the approach
High Precision Rotation Angle Estimation for
Rotated Images
17

18. Resolution Total Images Correct Images Correct Rate
10
500 486 97.2%
0.80
500 480 96.0%
0.60 500 471 94.2%
0.40 500 459 91.8%
0.20
500 438 87.6%
High Precision Rotation Angle Estimation for
Rotated Images
18

19.  Experiment results. 1st column: three images rotated at 50
, 250
, 450
respectively;
2nd column: theoretical pixel variance spectrum for the rotated images; 3rd
column: actual pixel variance spectrum for the rotated images.
High Precision Rotation Angle Estimation for
Rotated Images
19

20.  Conclusion
 In this paper, propose a blind image rotation angle estimation
method is proposed by exploring the periodicity of pixel variance
of rotated images.
 Experiment results show that this method works well for rotation
angles larger than 50
, but not as good for smaller rotation angles.
 The method can be used in areas like copy-paste image forgery
detection. In the future, the author plans to modify the algorithm
to improve the correct rate of small rotation angle estimation.
High Precision Rotation Angle Estimation for
Rotated Images
20

21. Region Duplication Forgery Detection
using Hybrid Wavelet Transforms
21

22. Region Duplication Forgery Detection using
Hybrid Wavelet Transforms
 Starts by dividing the M×N suspicious image into small overlapping blocks.
 This step is achieved by sliding a window of size B×B from the upper left
corner to the lower right corner one pixel each time.
 The quantized DCT coefficients are extracted from each block and used to
represent the features of these blocks.
 The quantized DCT coefficients are stored as one row in a matrix A of (M-B+1)
× (N-B+1) rows and B× B columns, where B× B is the block size.
 Two identical rows in the matrix A, correspond to two identical blocks in the
suspicious image.
Discrete Cosine Transforms
22

23. Region Duplication Forgery Detection
using Hybrid Wavelet Transforms
Hadamard Walsh Transforms
The Product of a Boolean Function and a Walsh Matrix is a Walsh Spectrum
23

24. Region Duplication Forgery Detection
using Hybrid Wavelet Transforms
 Example of Copy-Move Forgery, (a) Original Image (b) Forged Image
24

25. Thank you
25