multimedia forances explaniations, Schemes and Mind Maps of Multimedia Applications

Download multimedia forances explaniations and more Schemes and Mind Maps Multimedia Applications in PDF only on Docsity!

Digital Image Forensics

Outline

 Motivation

 Image Source Identification

 Source Camera-Model Identification

 Source Camera/Camcorder Matching

 Image Origin Determination

 Discrimination of synthetic images

 Tamper Detection

 Open problems

The Ever-Growing Problem

 Determining whether the image has undergone any

form of modification or processing after it has been

(originally) captured.

The photo on the front page of

the Los Angeles Times

Tourist of death

Giant Skeleton!

Stalin with and

without Yezhov

Digital Image Forensics

 Problem is deeper, many aspects

 Need techniques to uncover facts about the origin,

veracity and nature of digital images

 Exif header cannot be trusted

 Watermarking technologies are not adopted

Synthetic

Scanner

Image

Forensics

Tamper

Detection

Covert Channels

(Steganogprahy)

Source/Origin

Identification

Camera

Source Identification

1. Color Interpolation

2. Gamma Correction

3. Color Conversion

4. White point correction

5. Compression

Source Camera-Model

Identification

 Digital camera-models have unique characteristics

 Challenges

 Many brands use components by the same manufacturer

 Processing steps remain same or very similar among

different models of a brand

CFA Detector

Lens Processing Filters

Color Interpolation

Gamma Correction

White Point Correction

Color Conversion

Compression

Deployed Image Features

 A set of image characteristics are extracted

 Energy in sub-spectral bands

 Higher order statistics of sub-band coefficients

 Image quality metrics

 Inter-band correlations

 Gamma factor estimates

 Deviations from gray world assumption

 A multi-class SVM classifier is designed

based on above features

Digital Cameras

 Results

Nikon Sony Canon

S

Canon

S

Canon

S

Nikon %89.67 %0.22 %4.77 %1.64 %3.

Sony %3.56 %95.2 %0.31 %0.34 %0.

Canon S110 %7.85 %0.6 %78.71 %4.78 %8.

Canon S100 %3.14 %0.32 %3.57 %92.84 %0.

Canon S200 %5.96 %2.27 %7.88 %0.23 %83.

Sony vs Nikon Four-camera case

Feature 7

Feature 13

Feature 9

Feature 1 Feature 4

Feature 8

Feature 1: White point correction in the Red channel

Feature 4: Czekanowski similarity measure

Features 7-13: Energy in various spectral bands

Confusion tables for 4- and 5-camera cases

Fuji Canon Sony Nikon

Fuji %92.3 %6.6 %0 %1.

Canon %0 %93.5 %3.3 %2.

Sony %0 %1.1 %98.9 %

Nikon %0 %0 %2.1 %97.

Sample scatter plots for various image features

Demosaicing Artifacts

Source Camera-Model

Identification

Identification Method:

(Bayram et al.)

Detecting Demosaicing Artifacts

 Demosaicing operation introduces correlation between image

pixels

 In busy image parts proprietary and highly non-linear

 In smooth image parts linear and low-order interpolation

 The correlation pattern is periodic

 A multi-class classifier is deployed based on periodicity features

Detection Results

 Results based on combined metrics

Canon Datron Hp Sony

Canon 82 11 3 10

Datron 11 88 0 2

Hp 0 0 91 2

Sony 7 1 6 86

Confusion table for 5-camera case

Confusion table for 4-camera case

Canon Datron Hp Kodak Sony

Canon 74 11 7 2 6

Datron 8 78 8 1 0

Hp 10 6 66 7 6

Kodak 4 0 4 86 1

Sony 4 5 15 4 87

Accuracy %

Accuracy %

More on Demosaicing Artifacts

Source Camera-Model

Identification

Identification Method:

(Swaminathan et al.)

Radial Distortion

 Lens produces aberration in images due to design and

manufacturing process.

 The most severe of lens aberration is radial distortion

 Different type of lenses introduce different degrees of radial

distortion

 Most explicit in inexpensive wide-angle lenses.

 The radial distortion causes straight lines in the object space

rendered as curved lines on the film or camera sensor.

Undistorted Barrel distortion Pincushion distortion

Measuring Radial Distortion

 The degree of radial distortion can be used to link an

image to a lens and to a subset of digital cameras

 The radial distribution can be expressed as

 The distortion parameters can be estimated iteratively

 Sub-pixel edge detection is performed

 Polygonal approximation is applied to extract possibly

distorted line segments

 Distortion error is computed between the distorted line

segments and the corresponding straight lines.

 Parameters are optimized to minimize distortion error

undistorted radius distorted radius

First-order and second-order

distortion parameters