Digital image forgery detection
Download as PPTX, PDF11 likes11,182 views
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.
![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](https://image.slidesharecdn.com/digitalimageforgerydetection-151101142154-lva1-app6892/85/Digital-image-forgery-detection-15-320.jpg)
![ 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](https://image.slidesharecdn.com/digitalimageforgerydetection-151101142154-lva1-app6892/85/Digital-image-forgery-detection-16-320.jpg)
Digital image forgery detection
- 1. 1
- 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
Editor's Notes
- #5: Semantics is the subfield that is devoted to the study of meaning, as inherent at the levels of words, phrases, sentences, and larger units of discourse (termed texts, or narratives).
- #7: Montage (/mɒnˈtɑːʒ/) is a technique in film editing in which a series of short shots are edited into a sequence to condense space, time, and information. The term has been used in various contexts.