Image Forgery Identification Using JPEG Intrinsic Fingerprints

1
Image Forgery Identification Using JPEG Intrinsic
Fingerprints

• Group
• Ayichew Hailu
• Srinivasa Rangan
• Ashutosh Garg

Advisors Dr. Bernd Girod Dr. Min Wu David
Varodayan
2
Project Goals

• Develop algorithms to detect and classify
  possible tamperings
• Motivation Image validation
• Security, law enforcement intelligence
• Possible tampering
• Cropping, rotation, blurring, sharpening,
  insertion etc.

3
Assumptions

• One forgery per image
• Grayscale images
• Algorithms easily extendible to color images
• Restrict forgeries
• Helps achieve forgery classification
• Limited to copy-paste, rotation, cropping and
  brightness

4
Problem Approach

• Phase 1
• JPEG compress ? tamper ? save bitmap
• Q table available
• Develop forgery detection algorithms
• Phase 2
• Estimate Q table based on JPEG standard Annex K
• Phase 3
• JPEG compress ? tamper ? JPEG compress
• Estimate primary Q table

5
Forgery measure

• For each block
• Compute DCT Transform
• Obtain metric based on remainder w.r.t. entries
  of Q matrix
• where
• and gives the absolute difference between a and
  the closest multiple of b

6
Forgery measure cont.

• Sum across blocks to obtain measure for image
• Compare against threshold
• Obtained through training

7
Estimating Threshold
Effect of Threshold on the Performance of Method
8
Copy-Paste Forgery
9
Original

• Copy-Paste Forgery

Forgery
10
Forgery

• Copy-Paste Forgery

Result
11
Cropping

• Search 8x8 block to fit DCT grid
• if successful, identify as cropping
• Repeat on last block
• Exception
• cropping preserves DCT grid

12
Rotation

• Rotate by 90 check distortion
• If lt threshold, declare rotation forgery
• Test fails for 180 rotation

Brightness

• Checks for high level of brightness changes
• Saturation and threshold based

13
Estimating Q Matrix for Single Compressed JPEG
Image

• For each DCT coefficient
• Obtain histogram
• Quantization step Distance between successive
  peaks of histogram
• Compute power spectral density Ye, 2007
• Estimate first 3X3 coefficients of Q matrix
• Obtain other values by finding closest match to
  standard matrices using lookup table

14
Histogram Plots
15
Performance
16
Test Result
Cropping forgery
Copy-Paste forgery
Rotation forgery
False Positive Rate and False Negative Rate
versus Quality Factor
17
Double JPEG Compression

• Primary quantization matrix estimated
• Fridrich, 2003
• Reliable estimate if second QF is greater than
  first
• Estimate used for previous algorithms
• Successfully tested on limited images

18
Future Work

• Extend methods to classifying other kinds of
  forgeries.
• Add more features in Q matrix approach to
  improve accuracy

19
Conclusions

• Algorithm detects forgeries with good accuracy
  for both single and double JPEG compressed images
• Accuracy improves with higher quality factor
• Cropping and rotation are easiest to detect while
  copy-paste is hardest to detect

20

• Thank You

21
References

• 1 F. Fridrich, D. Soukal, and J. Lukas,
  Detection of Copy-Move Forgery in Digital
  Images, Digital Forensic Research Workshop,
  Cleveland, USA, Aug. 2003.
• 2 W. Luo, Z. Qu, J. Huang, G. Qiu, "A Novel
  Method for Detecting Cropped and Recompressed
  Image Block," Acoustics, Speech and Signal
  Processing, 2007. ICASSP 2007. IEEE International
  Conference on , vol.2, no., pp.II-217-II-220,
  15-20 April 2007.
• 3 J. Lukas and J. Fridrich, Estimation of
  Primary Quantization Matrix in Double Compressed
  JPEG Images, Proc. of the Digital Forensics
  Research Workshop, Cleveland, OH, Aug. 2003.
• 4 G. Schaefer, and M. Stich, UCID An
  Uncompressed Color Image Database, Technical
  Report, School of Computing and Mathematics,
  Nottingham Trent University, U.K., 2003.
• 5 A Swaminathan, M. Wu, K. Liu J. R., "Digital
  Image Forensics via Intrinsic Fingerprints,"
  Information Forensics and Security, IEEE
  Transactions on , vol.3, no.1, pp.101-117, March
  2008.
• 6 S. Ye, Q. Sun, E. and Chang, "Detecting
  Digital Image Forgeries by Measuring
  Inconsistencies of Blocking Artifact," Multimedia
  and Expo, 2007 IEEE International Conference on ,
  vol., no., pp.12-15, 2-5 July 2007.

22
Appendix
23
Rotation

• A rot90(X) ( X rot90(A) )T
• A rot180(X) AT
• ( rot90(X) rot90(A) )T AT
• rot90(A)T ( A rot90(X) )T
• rot90(A)T X rot90(A)