Overview of State-of-the-Art in Digital
Image Forensics——Image Source
Identification
Author: H. T. SENCAR and N. MEMON
Reporter: Yao Ge
1
Motivation 1/2
• In today’s digital age, the creation and manipulation of
digital images is made simple by low-cost hardware and
software tools.
• As a result, we are rapidly reaching a situation where one
can no longer take the authenticity and integrity of digital
images for granted (and more) .
2
Motivation 2/2
• The Image Forensics can help us to evaluate the reality
and integrity of a given digital image
• Eg. evidence
• One of the most popular research directions is Image
Source Identification (ISI)
What is Image Source Identification (ISI)?
• Purpose
– To identify the characteristics of digital data
acquisition device (e.g. Digital camera)
• Aspects:
– Source mode Identification
– Individual Source Identification
4
Image Acquisition Pipeline 1/3
Lens
Filter(s)
Color Filter
Array
Sensor
Camera
Processing
5
Image Acquisition Pipeline 2/3
• Lens system
– composed of a lens and the mechanisms to control exposure, focusing, and
image stabilization to collect and control the light from the scene.
• Filters
– includes the infra-red and anti-aliasing filters to ensure maximum visible quality.
• Image sensor
– An image sensor is an array of rows and columns of photodiode elements, or
pixels. When light strikes the pixel array, each pixel generates an analog signal
proportional to the intensity of light, which is then converted to digital signal and
processed by the DIP.
• CFA
– Since the sensor pixels are not sensitive to color, to produce a color image, a
color filter array (CFA) is used in front of the sensor so that each pixel records
the light intensity for a single color only.
Image Acquisition Pipeline 3/3
• DIP
– The output from the sensor with a Bayer (RGB) filter (assume) is a
mosaic of red, green and blue pixels of different intensities. Each pixel
contains the information of only one color. The digital image processor
implements interpolation (demosaicing) algorithms to recover the
missing information of the other two colors for each pixel.
– The DIP also performs further processing such as white balancing,
noise reduction, matrix manipulation, image sharpening, aperture
correction, and gamma correction to produce a good quality image.
Source Model Identification
• Image Features
• CFA and Demosaicing Artifacts
• Lens Distortions
Image Features 1/4
• Principle
– Features extracting + Classifiers (criterion)
training procedure
Images (Model 1)
features
Images (Model 2)
features
Images (Model 3)
features
classifer
testing procedure
classifer
Test Image (Model ?)
features
Model 1
Model 2
Model 3
9
Image Features 2/4
• E.g.
– Defines a set of 34 features inspired by universal steganalysis
techniques
– Features: Color features, wavelet coefficient statistics, image
quality metrics
*M. Kharrazi, H. T. Sencar, and N. Memon, Blind Source Camera
Identification, Proc. of IEEE ICIP (2004)
Image Features 3/4
• Experimental Results
– 2 cameras case
(average accuracy=98.73%)
– 5 cameras case
(average accuracy=88.02%)
*M. Kharrazi, H. T. Sencar, and N. Memon, Blind Source Camera
Identification, Proc. of IEEE ICIP (2004)
Image Features 4/4
• Weakness
– an overall decision. What specific features contribute much
during the process are still unknown
– this method may not give a satisfactory result with the increasing
of the number of cameras
• Conclusion
– this approach is more suitable as a pre-processing technique to
cluster images taken by cameras with similar components and
processing algorithms
*M. Kharrazi, H. T. Sencar, and N. Memon, Blind Source Camera
Identification, Proc. of IEEE ICIP (2004)
CFA and Demosaicing Artifacts 1/7
• Choice of CFA
Bayer Pattern (RGB)
CMYK
• Demosaicing
13
CFA and Demosaicing Artifacts 2/7
• Principle
– As the interpolation algorithms differ from each other in different
camera models.
• E.g.
– The author uses the outputs of Expectation/Maximization (EM)
algorithm as features to detect different interpolation.
S. Bayram, H. T. Sencar and N. Memon, Source Camera Identification
Based on CFA Interpolation, Proc. of IEEE ICIP (2005)
14
CFA and Demosaicing Artifacts 3/7
• Expectation/Maximization (EM) Algorithm
– EM algorithm is originally used to detect whether a signal has
been re-sampled or not.
– It generates two outputs
• The probability map. The value of each point on the probability map
indicates the probability that the point is correlated with its
neighbors.
• The estimate of the weighting coefficients which represent the
amount of contribution from each pixel in the interpolation kernel.
S. Bayram, H. T. Sencar and N. Memon, Source Camera Identification
Based on CFA Interpolation, Proc. of IEEE ICIP (2005)
CFA and Demosaicing Artifacts 4/7
• several examples of the periodic patterns that emerged
due to re-sampling
S. Bayram, H. T. Sencar and N. Memon, Source Camera Identification
Based on CFA Interpolation, Proc. of IEEE ICIP (2005)
CFA and Demosaicing Artifacts 5/7
• Frequency spectrum of probability maps obtained by
three types of digital cameras
S. Bayram, H. T. Sencar and N. Memon, Source Camera Identification
Based on CFA Interpolation, Proc. of IEEE ICIP (2005)
CFA and Demosaicing Artifacts 6/7
• Experimental results
– The set of weighting coefficients obtained from an image, and
the peak location and magnitudes in frequency spectrum are
used as features. An SVM classifier is used.
S. Bayram, H. T. Sencar and N. Memon, Source Camera Identification
Based on CFA Interpolation, Proc. of IEEE ICIP (2005)
CFA and Demosaicing Artifacts 7/7
S. Bayram, H. T. Sencar and N. Memon, Source Camera Identification
Based on CFA Interpolation, Proc. of IEEE ICIP (2005)
Lens Distortions
Radial distortion
Rectified Image
• Radial distortion is due to the change in the image
magnification with increasing distance from the
optical axis
• Compensation for radial distortion induces unique
artifacts in the images
• Choi et al. introduces a second order radial
symmetric distortion model
– Model parameters are used as classification features
– Accuracy ~91%
K. S. Choi, E. Y. Lam and K. K. Y. Wong, Source Camera Identification
Using Footprints from Lens Aberration, Proc. of SPIE (2006).
20
Conclusion
• Even the initial experiment results of
previous methods are encouraging, some
technique limitation still exist in different
real situation.
• There is a long distance to achieve the
application level.
Thank you!
Dec. 2008