Image Forgery Detection

Course Image Forensics

FCS482
By
Dr Mai Kamal El Den Mohamed
Lecturer, Computer Science Department,
Faculty of Computers and Artificial
Intelligence, Benha University
 Image Forensics

Introduction
Lecture 3
Image Forensics Techniques
Publication in the Field Digital Forgery Detection
 forgery detection in digital images has become a very popular research topic.
 The number of papers that addressed image forgery detection in IEEE and science
direct over last 10 years.
Image Forensics
 Digital Image forensics is an emerging branch of image
processing,
 Digital Image Forensics focuses on analyzing digital images to
determine their authenticity, detect manipulation, and retrieve
useful forensic evidence. It is gaining importance due to the
increasing cases of image forgery, deepfakes, and
misinformation in digital media.
 Image forensics techniques can be classified under two different
approaches, Active approaches and Passive/Blind approaches.
Classification
Classificationof
of Image Authentication
Image Forgery Techniques
Detection Techniques

Digital signature
Active
Digital
watermarking

Block-based
Image Copy-
Forgery move(cloning) Key-point
based(feature
Detection Statistical point matching)

techniques Pixel-based
Splicing
Format-based
Resampling
Passive(Blind) Image Forensics Camera based

Physically based

Geometric base
 Passive Authentication
 Passive authentication also called image forensics is the
process of authenticating images with no requirement of prior
information just the image itself.
 Passive techniques are based on the assumption that even
though tampering may not leave any visual trace but they are
likely to alter the underlying statistics.
 It is these inconsistencies that are used to detect the tampering.
 Active Authentication
 In active authentication techniques prior information about the
image is indispensable to the process of authentication. It is
concerned with data hiding where some code is embedded into
the image at the time of generation.
• Active techniques are based on inserting watermarks or
signatures in digital images during recording.
• The authenticity is determined by checking if the extracted
watermark from the image matches the original watermark.
Classification
Classificationof
of Image Authentication
Image Forgery Techniques
Detection Techniques

Digital signature
Active
Digital
watermarking

Block-based
Image Copy-
Forgery move(cloning) Key-point
based(feature
Detection Statistical point matching)

techniques Pixel-based
Splicing
Format-based
Resampling
Passive(Blind) Image Forensics Camera based

Physically based

Geometric base
 Digital Content Protection
• The most popular practices in this field include digital
content protection techniques such as:
1) watermarking and,
2) steganography.
Digital Content Protection
• Steganography takes a unique approach, concealing
information within digital content while remaining
undetectable.
• Digital watermarking is a powerful technique used to
embed additional information into digital content,
such as images, videos, audio, or documents, without
visibly altering the content.
• Forensic watermarking is a specialized subset of
digital watermarking, specifically designed to identify
the source of unauthorized distribution or leaks.
Steganography
Digital Watermarking
Digital Content Protection
Digital Watermarking
 Image Authentication Using Active Watermarking is a technique that
embeds a watermark (such as a digital signature, logo, or unique pattern)
into an image to verify its authenticity and detect any tampering.
 Digital image watermarking can be used to protect a copyright,
production, illegal distribution, unauthorized manipulation, piracy, forgery,
theft and ownership of digital image.
 Image watermarking technique embeds an authorized mark information
in the digital image to protect the ownership of digital image.
 Embedding a watermark without degrading the perceptual image quality
and at the same time making it difficult to remove.
 Embedding and extracting the watermark image should be limited to
authorized users only.
Active Image Forgery Detection
Techniques
Forensic Watermarking
Watermark Classification
 Forensic Watermarking
Human Perception
Visible Watermark
• A watermark logo can be viewed as the property of the ownership.
 Forensic Watermarking
Human Perception
InVisible Watermark
• Distortion of embedded watermark is too small to be noticed.
• A watermark can be extracted by authorized extraction software.

Watermark

Host image Watermarked image

Forensic Watermarking
Robustness
Fragile/ Semi Fragile/ Robust
• Watermark fails to be detected from watermarked image after
slightly modification. (Fragile).

• Watermark resists beginning transformation, while a watermark

fails to be detected after malignant transformation (Semi-fragile).

• Watermark resists against different types of attack (Robust).

 Watermark Classification
Watermarking System
Watermark W
Watermark
Host Image I embedding Watermarked Image
Key K system
Watermark embedding process
Watermarked Image
Watermark
Test Data Extraction Watermark extraction
system
Key K
Watermark recovery process

 A watermarking scheme is a structured method of embedding

and retrieving a watermark in an image to verify authenticity,
prevent tampering, and protect intellectual property.
 The ideal watermarking is to embed such amount of information into host
image that provides tradeoff between robustness and imperceptibility by
human visual system.
Forensic Watermarking
Watermarking Techniques

Spatial Domain
Digital watermarking manipulates the pixel values for adding
watermark information.
For example: Least significant bit (LSB)

Frequency Domain
Adding or changing component of digital images can be
transferred to other value domains.
For example: Spread Spectrum Watermarking(SSW)
Spatial Domain
Least Significant Bit
• A true color image consists of three channels : red, green and blue
component. For each pixel has 1 byte / 8 bits, given a bright purple
as shown below:
Y0 = {R=254, G=1, B=254}
Least significant bit is a simple technique, embeds a watermark into
the least of bits. The result of watermark embedding based on LSB
in Y0 is given as:
Y1 = {R=254, G=1, B=253}
• Human visual system is almost impossible to detect a difference of 1
on a color scale.
• While the watermark can easily be removed when the watermarked
image is compressed by JPEG compression.
Spatial Domain
Least Significant Bit

• Random numbers based on a secret key can distribute watermark

embedding into the host image.
• Increasing the number of bits can give significantly effect to the
quality of watermarked image.
 Spatial Domain
Least Significant Bit
• Example Image Pixel Values (Grayscale)
– Assume we have a 2×2 grayscale image with 8-bit pixel values:

– Assume we want to embed the binary watermark: 1010.

• Sol.
– The corresponding binary representation of these pixel values:

– We replace the least significant bit of each pixel with the watermark
bits, and the New Image after Watermarking:
Frequency-Based Techniques
• Embedding watermark schemes in frequency coefficients will
produce higher robustness than embedding watermark schemes in
spatial domain.
• A watermarking scheme which uses frequency-based should embed a
watermark in low frequency order or medium frequencies.
• Embedding a watermark in a low frequency can produce a higher
quality degradation of the watermarked image.
• Embedding a watermark in a high frequency makes less robust and it
is higher probability being lost when the watermarked images are
compressed.
Frequancy Domain
Water mark
1. Discrete Cosine Transform (DCT) Based Watermarking
– DCT transforms an image into frequency components, dividing it
into low, middle, and high-frequency bands. Watermarking is
usually done in the middle-frequency bands to maintain balance
between robustness and imperceptibility.
2. Discrete Wavelet Transform (DWT) Based Watermarking
– DWT decomposes an image into sub-bands (LL, LH, HL, HH),
where:
– LL (Low-Low) represents approximation details (most important
visual data).
– LH, HL, HH capture edge and texture details.
3. Discrete Fourier Transform (DFT) Based Watermarking
– DFT represents an image in the frequency domain, where
modifications are more resistant to geometric attacks like rotation
and scaling.
Discrete Cosine Transform (DCT)
Based Watermarking
DCT B of an input image A is defined as:
M 1 N 1
 (2 m 1) p  (2 n 1) q
B pq   p  q   Amn cos cos ,
m 0 n 0
2M 2N
for p = 0, 1, 2, …, M1 and q = 0, 1, 2, …, N1
where  1  1
, for p0
, for  q0
 
p   M q   N
 2 , for  2 , for q0
p0 
 M  N

Inverse DCT is calculated by:

M 1 N 1
 (2 m 1) p  (2 n 1) q
A pq     p q Bmn cos
2M
cos
2N
,
m 0 n 0
Discrete Cosine Transform (DCT)
Based Watermarking
Discrete Wavelet Transform (DWT)
Based Watermarking
Discrete wavelet transform (DWT)
• The DWT has been widely implemented in the digital image
watermarking because of its multiresolution capability.
• DWT is a transform to decompose an image into the frequency bands
LL, LH, HL, and HH.
• The DWT technique performs multi-resolution and excellent
localization characteristics.
Discrete Wavelet Transform (DWT)
Based Watermarking
 Discrete Wavelet Transform (DWT)
Based Watermarking

DWT
Host images Multilevel DWT Coefficients
Fusion and
LL HL
IDWT rules
LH HH
Watermarked
Watermark images
1st level DWT
Key
Discrete Wavelet Transform (DWT)
Based Watermarking Example
• A low-pass filter (average operation) smooths the
values (reducing high variations).
• A high-pass filter (difference operation) extracts the
details (edges).
• The Watermark equation that used to apply modification:
• W′=W+α×Watermark
• Where:
– W' is the modified (watermarked) sub-band.
– W is the original sub-band (LL, LH, or HL).
– α is a scaling factor (controls watermark strength).
– Watermark is a matrix of the same size.
Discrete Wavelet Transform (DWT)
Based Watermarking Example
Let:
Discrete Wavelet Transform (DWT)
Based Watermarking Example
Discrete Wavelet Transform (DWT)
Based Watermarking Example
104 99
• 𝐿𝐿 =
114 106
−3 −1
• H𝐿 =
−1 −1
−1 1
• 𝐿𝐻 =
−3 2
0 0
• 𝐻𝐻 =
0 0
Discrete Wavelet Transform (DWT)
Based Watermarking Example
• Final Watermarked Sub-Bands
Spread-Spectrum Techniques
Spread-Spectrum Techniques
• Step 1: Define the Original Signal (Host Data).
• Step 2: Generate a Pseudo-Random Noise Sequence.
• Step 3: Embed the Watermark.
• Example:
• Assume we have a 1D signal (this could represent an image's
frequency coefficients):
• Let’s a watermark bit W=1.
• The watermark strength α (scaling factor) determines visibility.
We choose α=2.
• a binary pseudo-random sequence (PN) to spread the
watermark:
Spread-Spectrum Techniques
• Example:
 Foundations of Attacking

Three effects make Watermark cannot be detected.

detection of
watermarking useless:

False watermarks are detected.

Unauthorized detection of watermark.

Digital Watermark Drawbacks
• Digital cameras must be equipped with a watermark
insertion mechanism during recording, which would limit
the use of this technique with special cameras.
• Moreover, the watermarking technique may be
unreliable.
• A watermark could be extracted to manipulate the image
and then be reinserted.
Digital Watermark Drawbacks
• watermarks can be weak, and when a watermarked image
has been compressed using a compressing algorithm like
JPEG, they are destroyed.
• it is not possible to decide if watermark has been inserted
after image manipulation or not.
• This led to the need for the development of 26 passive or
blind techniques
Classification
Classificationof
of Image Authentication
Image Forgery Techniques
Detection Techniques

Digital signature
Active
Digital
watermarking √
Block-based
Image Copy-
Forgery move(cloning) Key-point
based(feature
Detection Statistical point matching)

techniques Pixel-based
Splicing
Format-based
Resampling
Passive(Blind) Image Forensics Camera based

Physically based

Geometric base
Digital Signature as an Active
Technique in Image Forensics

• A digital signature is an active image forensics

technique used to verify the authenticity and
integrity of an image. It works by generating a
unique cryptographic signature based on the
image's content, which can later be used to detect
modifications or tampering.
How Digital Signatures Work in
Image Forensics
1) Signature Generation
• A hash function processes the original image to produce a unique hash value
(a fixed-length digital fingerprint).
• This hash is then encrypted using the sender’s private key to create the digital
signature.
• The signature is either embedded in the image metadata or stored in a
separate database.

2) Verification Process
• When verifying an image, the same hash function is applied to the received
image.
• The stored digital signature is decrypted using the sender’s public key to obtain
the original hash.
• If the computed hash matches the decrypted hash, the image is authentic;
otherwise, it has been altered or tampered with.