Digital Forensics Ya l d a M o h s e n za d e h 30 Aban 1387
Outline 2
Introduction
Digital Forensics
Forensics & Watermarking
Applications
Nonintrusive Forensics Blind Identification of Source Camera Model Conclusion شاخه داوشجویی اوجمه رمز ایران The 1st Workshop on Info. Hiding
13:38
Introduction 3
شاخه داوشجویی اوجمه رمز ایران The 1st Workshop on Info. Hiding
13:38
Digital Forensics 4
High popularity of digital images
The integrity of image content ??? Low-cost and sophisticated image editing software
How an image was acquired? Has the image undergone any manipulation after capture? Does it contain any hidden information? شاخه داوشجویی اوجمه رمز ایران The 1st Workshop on Info. Hiding
13:38
Forensics & Watermarking 5
Data authentication techniques Watermarking Robust hashing Inserting watermark/signature at the time of creation of
multimedia data The presence or absence of the watermark
The authenticity of digital images
شاخه داوشجویی اوجمه رمز ایران The 1st Workshop on Info. Hiding
13:38
Forensics & Watermarking 6
Applicability restrictions for adding a watermark or a hash . A strong motivation to devise nonintrusive methods Modeling watermarking and steganography as postprocessing
operations and identifying them with such methods Universal blind steganalysis & nonintrusive forensics methods
شاخه داوشجویی اوجمه رمز ایران The 1st Workshop on Info. Hiding
13:38
Nonintrusive Forensics Categories 7
Source authentication Higher order statistical models using wavelet transform coefficients Physics-motivated features based on geometry features Tampering detection Extracting certain salient features to distinguish tampering such as:
JPEG compression Resampling Gamma correction, etc
Exhaustive search
Steganalysis شاخه داوشجویی اوجمه رمز ایران The 1st Workshop on Info. Hiding
13:38
Applications 8
Fingerprint scanner detects the living from the dead Iris recognition
It is not genetically determined. Stable throughout life
Highly complex and unique Camera identification شاخه داوشجویی اوجمه رمز ایران The 1st Workshop on Info. Hiding
13:38
Blind Identification of Source Camera Model 9
شاخه داوشجویی اوجمه رمز ایران The 1st Workshop on Info. Hiding
13:38
Image-Acquisition Model 10
S ( x, y, c) S p ( x, y, c) 0
p( x, y) c oth.
شاخه داوشجویی اوجمه رمز ایران The 1st Workshop on Info. Hiding
13:38
Camera Identification 11
Sensor noise
Demosaicking artifacts
Lens radial distortion
Sensor dust characteristics
Inter-pixel correlation model شاخه داوشجویی اوجمه رمز ایران The 1st Workshop on Info. Hiding
13:38
Sensor Noise 12
Fixed pattern noise of the CCD (Charge Coupled Device)
array A unique pattern for each camera A more persistent feature: The photo-response non-uniformity noise (PRNU) caused by pixel nonuniformities It acts like a spread spectrum watermark unique to each camera شاخه داوشجویی اوجمه رمز ایران The 1st Workshop on Info. Hiding
13:38
Demosaicing Artifacts 13
Commercial imaging devices use a single mosaic structured color filter array
(CFA). Camera models employ their proprietary interpolation algorithm in recreating
the missing color values. Foot print: Correlation patterns between contiguous bit planes
Features:
Mean value of RGB channels Correlations between color components Wavelet domain statistics
Inter-Pixel Correlation Model (PCA, Neural networks, sensitive to
median filtering) شاخه داوشجویی اوجمه رمز ایران The 1st Workshop on Info. Hiding
13:38
Alternative Methods 14
Lens radial distortion
Inherent radial distortions vary from one manufacturer to another SVM classification Influenced by focal length of the lens
Sensor dust characteristics
Using dust model of digital single lens reflex (DSLR) cameras Stable dust pattern Dust specks are hardly detected in non-smooth and complex regions of the image
A new method
Statistical moments of the image denoising residuals Camera model and brand discrimination
The 1st Workshop on Info. Hiding
شاخه داوشجویی اوجمه رمز ایران 13:38
SOURCE CAMERA IDENTIFICATION ALGORITHM 15
Test Image
Feature extraction
Scoring with SVM classifier
Fusion Techniques
Final Decision
شاخه داوشجویی اوجمه رمز ایران The 1st Workshop on Info. Hiding
13:38
Forensic Feature Types for Cameras 16
Characteristics of the lower order bit planes The occurrence of bit patterns within 3x3 neighborhoods Histograms in 3 dimension Spatial: Occur within a bit plane Quantal: Occur between two adjacent bit planes Chromatic: Occur across color components
histograms of manipulated images histograms of different Cell-phone models The 1st Workshop on Info. Hiding
شاخه داوشجویی اوجمه رمز ایران 13:38
Forensic Feature Types for Cameras 16
Characteristics of the lower order bit planes Characteristics of image denoising residuals Characteristics of independent correlationparts within and between wavelet Extracting content of the image Image Quality Measures bands High-Order Wavelet Statistics
mean, variance, skewness and kurtosis
شاخه داوشجویی اوجمه رمز ایران The 1st Workshop on Info. Hiding
13:38
Feature Selection and Classification 17
For optimizing the number of features Sequential Forward Feature Selection (SFFS) algorithm Analyzing the features in ensembles Constructing the final set by adding to and/or removing from the current set of features Until no more performance improvement is possible Support Vector Machine classifier شاخه داوشجویی اوجمه رمز ایران The 1st Workshop on Info. Hiding
13:38
Decision Fusion Method 18
Three feature categories Binary Similarity Measures Image Quality Measures High-Order Wavelet Statistics
3 Multiclass Classifiers
Decision fusion methods Confidence-level fusion
The probability that the given image belongs to that class
Rank-level fusion Abstract-level fusion شاخه داوشجویی اوجمه رمز ایران
The 1st Workshop on Info. Hiding
13:38
COMPARISON OF DIFFERENT DECISION-LEVEL FUSION SCHEMES 19
شاخه داوشجویی اوجمه رمز ایران The 1st Workshop on Info. Hiding
13:38
PERFORMANCE COMPARISON 20
شاخه داوشجویی اوجمه رمز ایران The 1st Workshop on Info. Hiding
13:38
Conclusion 21
Digital Forensics Digital forensics with watermarking system It is possible to identify the source cellular phone camera model
of an image. The camera identification scheme is robust against image
manipulations.
شاخه داوشجویی اوجمه رمز ایران The 1st Workshop on Info. Hiding
13:38
References 22
[1] O. Celiktutan, B. Sankur, I. Avcıbas, “Blind Identification of Source Cell-phone Model”, JOURNAL OF INFORMATION FORENSICS AND SECURITY, 2007 [2]A. Swaminathan, M. Wu, K. J. Ray Liu, “Digital Image Forensics via Intrinsic Fingerprints”, IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY, VOL. 3, NO. 1, MARCH 2008 [3] B. Park, A. Savoldi, P. Gubian, J. Park, S. H. Lee, S. Lee, “Recovery of Damaged Compressed Files for Digital Forensic Purposes “, International Conference on Multimedia and Ubiquitous Engineering 2008 [4] Y. Ashino, R. Sasaki, “Proposal of Digital Forensic System Using Security Device and Hysteresis Signature”, IEEE 2007 شاخه داوشجویی اوجمه رمز ایران The 1st Workshop on Info. Hiding
13:38
The 1st Workshop on Info. Hiding
23
13:38