Multimedia Tools and Applications

https://doi.org/10.1007/s11042-023-18090-y

A survey on digital image forensic methods based on blind

forgery detection

Deependra Kumar Shukla1 · Abhishek Bansal2 · Pawan Singh3

Received: 9 October 2022 / Revised: 26 September 2023 / Accepted: 29 December 2023

© The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2024

Abstract
In the current digital era, images have become one of the key channels for communication
and information. There are multiple platforms where digital images are used as an essential
identity, like social media platforms, chat applications, electronic and print media, medical
science, forensics and criminal investigation, the court of law, and many more. Alternation
of digital images becomes easy because multiple image editing software applications are
accessible freely on the internet. These modified images can create severe problems in the
field where the correctness of the image is essential. In such situations, the authenticity of
the digital images from the bare eye is almost impossible. To prove the validity of the digi-
tal images, we have only one option: Digital Image Forensics (DIF). This study reviewed
various image forgery and image forgery detection methods based on blind forgery detec-
tion techniques mainly. We describe the essential components of these approaches, as well
as the datasets used to train and verify them. Performance analysis of these methods on
various metrics is also discussed here.

Keywords Forgery · Copy-move · Cloning · Imaging device · Digital forensic · Deep-

learning

* Pawan Singh
[email protected]
Deependra Kumar Shukla
[email protected]
Abhishek Bansal
[email protected]
1
Department of Computer Science, Indira Gandhi National Tribal University, Amarkantak,
Madhya Pradesh 484887, India
2
Department of Computer Science & Applications, Dr. Harisingh Gour Vishwavidyalaya, Sagar,
Madhya Pradesh 470003, India
3
Department of Computer Science, Central University of Rajasthan, Ajmer, Rajasthan 305817,
India

13
Vol.:(0123456789)
 Multimedia Tools and Applications

Fig. 1  Different image forgery methods

(a) Genuine image. (b) Image after retouching.

Fig. 2  Retouching image forgery [13]

1 Introduction

The term "digital image forgery" refers to the process of creating fake or altered images.
Altering images is now effortless due to publicly available powerful image editing tools
like Corel PaintShop Pro, Affinity Photo, ACDsee photo editor, and Adobe Photoshop
[1–6]. We have used a systematic literature review methodology to write this review
paper. We have identified and critically appraised relevant research for collecting and
analysing data for blind forgery detection techniques.
Digital image fraud has occurred in several ways. On the basis of techniques used to
generate fake images, all of these examples can be categorised into seven categories:
Splicing, Retouching, Morphing, Copy-Move or Cloning, Inpainting, Resampling, and
Deep-fakes [7–10]. Figure 1 demonstrates the various types of image forgery methods:

1.1 Image retouching

Maximizing or minimizing any particular feature of the image is called image retouch-
ing. This technique is less harmful and is commonly used by the magazine’s photo edi-
tors. Most probably, all the magazine cover photos use this technique to enhance any
particular image aspect to make it more attractive. At the same time, such modifications
are unethical [10–12]. Figure 2 is an excellent sample of retouching forgery. Sub-image

13
Multimedia Tools and Applications

(a) Genuine images (b) Spliced image

Fig. 3  Image splicing forgery [17]

(a) Original image (b) Morphed image (c) Original image

Fig. 4  Image morphing Forgery [21]

(a) is an authentic image, and sub-image (b) is an altered image created by retouching
[13].

1.2 Image splicing

In image splicing, an altered image is created using the different portions of two or more
images. To make spliced images, simple cut and paste method is used. This technique of
creating fake images is very popular among attackers. These spliced images can be propa-
gated on social media with bad intentions. Therefore, we can say that this method can cre-
ate more trouble than image retouching [10, 11, 14–16]. Figure 3 demonstrates the image
splicing.
Here, sub-image (a) is a genuine image, and sub-image (b) is the spliced version of the
genuine image.

1.3 Morphing

The transformation of one image into another with the help of intermediate images is
called image morphing. The goal is to create a series of intermediate images that indicate
the transition from one image to another when combined with the originals [18–20]. Fig-
ure 4 shows the image morphing forgery. Sub-image (a) and sub-image (c) are the authen-
tic images, whereas sub-image (b) is the morphed image of (i) and (iii) images [21].

13
 Multimedia Tools and Applications

(a) Authentic image. (b) Image created by a copy-move attack

Fig. 5  Copy- Move image forgery [25]

(a) genuine image (b) Masked image (c) Inpainted image

Fig. 6  Image Inpainting Forgery[28]

1.4 Copy‑Move or Cloning attack

In the copy-move attack, the attacker copied from one region of the image and pasted that
onto other regions of the same image to remove any specific portion from the image. This
attack is used to create forged images [4, 8, 10, 11, 15, 22–24]. Figure 5 demonstrates the
copy-move attack in digital photographs. Sub-image (a) is the authentic image, whereas
sub-image (b) is the altered image created by a copy-move attack [25].

1.5 Image Inpainting

Generally, image inpainting is used to fill blank-spaces or corrupted regions whitin the
image. Unfortunately, attackers used this approach to conceal a specific portion of the
image with malicious intentions to create a fake image [26, 27]. Figure 6 demonstrate an
excellent example of image inpainting. Figure 6 has three sections a, b, and c. Section a

13
Multimedia Tools and Applications

contains two original images, section b contains the masked image of images in section
a, then finally inpainted images corresponding to masked images are created in section
c.

1.6 Resampling attack

Resampling involves generating a modified version of the original image through opera-
tions like stretching, rotation, and scaling [9]. To fraudulently geometrically change a
digital image or a piece of a digital image, the resampling technique uses interpolation
algorithms [29]. Figure 7 shows the Resampling forgery with flipping, scaling, and rota-
tion operation.
Here in the Fig. 7, sub-image (a) is the authentic image, sub-image (b) is the resam-
pled image by using the flipping operation, sub-image (c) is the resampled image with
scaling, and sub-image (d) is the resampled image with the scaling and rotation opera-
tion [30].

1.7 Deep‑fake images

Other than above mentioned traditional ways of forging images, currently new meth-
ods have been developed to create a fake image using Artificial Intelligence techniques
called deep-fake. Deepfake images are computer-generated or altered pictures produced
through advanced machine learning methods like Generative Adversarial Networks
(GANs). These images convincingly depict individuals, objects, or environments that
either never existed or have been substantially modified [27, 31].

Fig. 7  Resampling Image forgery [30]

13
 Multimedia Tools and Applications

2 Digital image forensic

We have seen various ways to forge an image and destroy its integrity. So here, one
crucial question arrives how to prove the image’s purity and reliability? The answer to
this question is Digital image forensics (DIF). DIF is the most recent field of research
that solves two fundamental issues: imaging device identification and forgery or tamper
detection [32]. In DIF, two methods are available to detect the alternation in the altered
images. The first approach is the Active technique, and the second approach is the Inac-
tive or Blind or Passive technique [33–36]. Figure 8 shows the different forgery detec-
tion techniques.

2.1 Active techniques

These techniques need previous knowledge about the image elements linked to the
image, such as information related to steganography, water-marking or digital signature
[8, 9, 27, 37, 38].

2.1.1 Steganography

In steganography, confidential data like text, images, or videos are hidden using the
cover image. To achieve steganography, steganographic techniques are of two catego-
ries. The first is frequency domain methods, and the second is spatial domain methods
[39–41].

2.1.2 Water‑marking

Inserting a piece of confidential information in the digital image to preserve its integrity
is known as water-marking. A water-mark in an image can be either visible or invisible
[42, 43].

Fig. 8  Techniques of image fraud detection

13
Multimedia Tools and Applications

2.1.3 Digital signature

Digital signatures can also be inserted into the image during its creation to preserve the
purity of the image. Digital signatures can be extracted and compared to confirm the
image’s authenticity [10].

2.2 Passive Techniques

This technique does not require any previous knowledge related to image; therefore, this
technique is also known as the blind technique [8, 9, 27, 37, 38]. Our primary focus in
this survey is to study passive or blind forgery detection methods. Passive forgery detec-
tion includes Imaging device identification, copy-move detection, and splicing detec-
tion. Figure 9 demonstrate the different components of blind forgery detection methods
[44].
All possible steps shown in Fig. 9 are widely explained in the following section.

Input: Suspicious images will be given as input to detect possible blind forgery [45].
Pre-processing: It is an optional step. Some images required it, and some did not. Some
pre-processing operations are given as RGB to Gray-scale, RGB to YCbCr, RGB to
HSV, Image resizing, etc. [45].
Feature extraction: Extraction of image features is essential for passive forgery detec-
tion. Some methods to extract features from the digital image are DCT, DWT, FMT,
Zernike moments, SIFT, SURF, CNN, ResNet50, VGG19, contourlet transform, etc.
[45].
Feature matching: Feature matching is employed to identify duplicated regions, where
a strong resemblance between two feature descriptors serves as an indicator of such
regions. Methods used for matching are PatchMatch, LSH, Lexicographical sorting,
g2NN, k-d tree, Counting BloomFilters, etc. [45].
Filtering: This step is applied to minimize the probability of false matching [45].
Post-processing: Post-processing is applied to remove the false detection [45].
Output: In this step input image is classified as forged or genuine image.

2.2.1 Imaging device identification

Imaging device identification techniques recognize the inherent evidence left in the
images by the corresponding imaging device or source camera used in the acquisition
of the image [8, 46]. Imaging device identification techniques examine the footprints
left behind during the acquisition of the image to frame a fingerprint that can recognize
the imaging device or camera model. In particular, source camera authentication can
offer helpful technological support for judicial verification. Imaging device identifica-
tion plays a significant role in enhancing the security of photographs, settling copyright
issues, avoiding false publicity, and combatting cybercrime along with compensation
claims, child exploitation, and criminal proceedings [47]. This survey paper presents the
performance of different methods in relation to precision, recall, F-1 score, and accu-
racy. Apart from these metrics we have provided details of other metrics like FPR, FNR,
and TNR bot not utilized in this paper. These performance parameters are based on True

13
 Multimedia Tools and Applications

Fig. 9  Components of blind forgery detection

positive, True negative, False positive, and False negative. According to [22, 48] True
positive, True negative, False positive, and False negative are defined as follows.

True positive ­(Tp): True positive means the number of correctly identified forgeries.
True negative (­ Tn): True negative means the number of correctly identified non-forgery.
False positive ­(Fp): False positive means the number of wrongly detected forgeries.
False negative (­ Fn): False negative means the number of un-detected forgeries.

Precision Precision shows us the percentage of correct positive predictions made. Preci-
sion can be calculated by using the Eq. (1) in terms of % [22, 48, 49]:

13
Multimedia Tools and Applications

Tp
Precision = ∗ 100% (1)
Tp + Fp

Recall Recall shows us the percentage of positive predictions made out of all positive pre-
dictions. Recall can be calculated by using the Eq. (2) in terms of % [22, 48, 49]:
Tp
Recall = ∗ 100% (2)
Tp + Fn

F‑1 Score The harmonic mean of precision and recall is the F-1 score. F-1 score can be
calculated by using the Eq. (3) in terms of % [22, 48, 49]:
2 ∗ Precision ∗ Recall
F-1 score = ∗ 100% (3)
Precision + Recall
Accuracy Accuracy shows us the percentage of correct predictions out of all predictions.
Precision can be calculated by using the Eq. (4) in terms of % [22, 48]:
Tp + Tn
Accuracy = ∗ 100% (4)
Tp + Tn + Fp + Fn

False positive rate False positive rate is given by Eq. (5) in terms of % [50–52].
Fp
Falsepositiverate(FPR) = ∗ 100% (5)
Tn + Fp

False negative rate False negative rate is given by Eq. (6) in terms of % [50, 51].
Fn
Falsenegativerate(FNR) =
Tp + Fn
∗ 100% (6)

True negative rate True negative rate is given by the Eq. (7) in terms of % [51].
Tn
Truenegativerate(TNR) =
Fp + Tn
∗ 100% (7)

Figure 10 shows the primary methods used in imaging device identification.

i. Photo-response-noise-uniformity (PRNU) based methods

Fig. 10  Imaging device identification techniques

13
 Multimedia Tools and Applications

PRNU is used to recognize the corresponding imaging device from which the image
is taken. PRNU noise pattern is extracted from the input image [53]. Here we have pro-
vided two PRNU-based imaging device identification methods.

• [54] Proposes a model to identify imaging devices based on PRNU extraction

using a denoising filter. From acquired data like photographs or videos, the “Photo
Response Noise Uniformity” is a unique pattern that can be utilized to identify the
model of the source imaging devices. A reference pattern or fingerprint can be gen-
erated by averaging the obtained PRNU patterns from the image. The query image
from the testing data is accurately classified after analysis of all the reference pat-
terns of 25 devices by comparing the correlation values. The developed model is
tested using 125 photos form Dresden Image Database. Since the correlation value
is determined statistically, the accuracy is 100%. The suggested solution is tested on
limited camera models and their images (25 camera models and 125 images) there-
fore accuracy of the suggested approach must be examined over bigger datasets.
Robustness of the model’s performance should also be examined over various image
attacks.
• [55] Proposes a new approach that concentrates on the pixel that participates in sensor
noise in the PRNU pattern to distinguish imaging devices (cameras) of the same type.
The Jaccard coefficient is used to calculate the proportion of similar pixel locations
shared between two devices’ noise patterns. Their test findings demonstrate that using
cameras of the same kind from the Apple iPhone 6, FujiFilm, Panasonic, and Sony
cameras, the proposed system can correctly identify the device origin of the image. The
suggested approach accurately distinguishes between two Apple iPhone 6 devices that
are identical with 100% precision, With 100% accuracy, the FujiFilmXTlO, 97.50%
accuracy for the Panasonic DMC-FZlOOO, and 98.73% accuracy for the Sony ILCE-
6000. The Performance of the proposed solution is tested on only four camera models
therefore system performance must be examined on more camera models. Apart from
this, suggested solution can be extended for Source imaging device identification when
image is taken from similar type of imaging device.

ii. Feature extraction-based methods

Feature extraction-based techniques uses the extracted feature from the image to iden-
tify the imaging device. Here we have provided a study that uses the extracted feature from
the image to identify the imaging device. [56] Proposed a method for identifying the source
camera using the texture features of the image, which are taken from selected color models
and color channels. They separately extracted local binary pattern (LBP) features, the coef-
ficient of the contourlet transform, and residual noise images from the original image in
the recommended work. They also separately extract features of Local phase quantization
(LPQ) from the residual and authentic images. Hue saturation value (HSV) color space is
used to extract the LBP and LPQ characteristics. Then, they merge LPQ and LBP as dis-
tinct features to identify the camera source and input them into a “Multi-class Lib-SVM”
classifier. The following are the primary benefits of the suggested method:

• Images taken with the same brand and model of imaging devices can nevertheless be
distinguished from one another.
• It can withstand treatments that preserve content or geometric alternations such as scal-
ing, rotation, and JEPG compression.

13
Multimedia Tools and Applications

The experimental outcomes demonstrate that the suggested strategy achieves 99.76% of
detection accuracy. The proposed model used the 300 images for training and 300 images
for testing, therefore its performance must be examined over the bigger datasets. Apart
from this suggested solution uses the fixed size images therefore this work can be extended
for the variable size images.
iii. Camera response function-based methods
The camera response function is used to relate the scene radiance to the intensity
value[57]. Here we have provided a study in which CRF is used to identify imaging
devices. According to the literature [58], when luminosity occurs at a sensor, the sensor
associates a pixel with the intensity value to the associated pixel in the image by camera
response functions (CRFs). The nonlinearity of CRFs impacts a few techniques like de-
blurring, photometric stereo, etc. The proposed work concluded in three remarks: Firstly,
radiometric calibration is made simpler with the help of decreasing the number of basis
vectors while preserving the same level of fit. Secondly, if CRFs can be predicted with
high enough accuracy, they may still be useful for forensic purposes despite having less
diversity. Finally, it demonstrated that the requirement to calibrate the imaging device prior
to de-blurring radiometrically is substantially loosened in that ringing effects might fre-
quently be avoided using an average CRF. In this study, they performed experiments and
successfully differentiated with an average accuracy of 96% to distinguish between a film
CRF and a digital camera CRF. The suggested solution uses a variety of digital camera
models but the models performance can be examined by including the mobile camera’s in
the dataset (Table 1). Otherthan this one can aim to examine camera modes that employ
tone-mapping techniques using specific pixel patch analysis.

2.2.2 Copy‑move detection

The primary objective of these techniques is to identify the areas in an image where a
copied part of the same image is pasted to detect the forgery [8, 10, 18]. The attacker can
also alter the copied region. Hence, the primary objective of a copy-move operation is to
remove or hide the elements of the picture to remove its legitimacy. After a copy-move
process, the altered images could then go through further processing, including rescaling,
filtering, and noise dispersion to hide the signs of forgery [8]. Figure 11 represents the
methods of copy-move fraud detection.

a) Key-points based methods

Methods based on Key-Point to detect copy-move fraud bring out features or key points
from the altered images, such as SIFT, geometric transformation, or robust noise. Key-
point-based methods have a better time complexity than block-based methods[59].

• Scale-invariant feature transform (SIFT) is a key-point-based method. [60] proposed

a new scheme to detect copy-move fraud based on key-points. This study used the
SIFT method for key-point extraction because of the SIFT algorithm’s rotation and
scaling invariance characteristics. Sometimes SIFT algorithm is unable to bring out
enough features from the smooth region. To overcome this drawback, they parti-
tioned the image into texture as well as smooth areas. Using the SIFT algorithm,
they bring out the features only from the texture regions whereas feature from the

13
13
Table 1  Summarizes the work related to imaging device identification
Title of the paper with citation Publication year Method used Explanation of used method proposed scheme Performance

“Source Camera Identification using Photo 2019 PRNU This method uses the noise created by the Accuracy = 100%
Response Noise Uniformity” [54] imaging sensors during the process of
acquisition which helps to construct a
fingerprint or signature that identifies the
model of the imaging device
“PRNU-based Source Camera Identification 2021 PRNU In the Proposed work, The Jaccard coef- Accuracy = 100% for two apple iPhone 6,
for Multimedia Forensics” [55] ficient is used to calculate the proportion 100% for FujiFilmXTlO,97.50% for
of similar pixel locations shared between DMC-FZlOOO,and 98.73% for Sony
two devices’ noise patterns ILCE-6000
“Source camera identification from image 2016 Feature Extraction These techniques use a set of attributes Accuracy = 99.76%
texture features” [56] brought out from an image to identify
signs in images captured with the same
sensor
“Analyzing Modern Camera Response 2019 CRF The sensor collects luminance in a non- Average Accuracy = 96%
Functions” [58] linear fashion, and the imaging device’s
signatures are made using information
associated with this relationship
Multimedia Tools and Applications
Multimedia Tools and Applications

Fig. 11  Various methods for the detection of copy-move fraud

smooth regions were extracted using the Fourier Mellin Transform (FMT). The sug-
gested models achieves 96.97% F-score. The suggested solution can be extended for
detecting copy-move frauds in videos.
• SURF(Speeded-Up-Robust-Transform) is an efficient as well as a reliable technique
for local representation of similarity-invariant and comparison of pictures. [61] used
the SURF method in the proposed work for key-point identification and extraction,
showing its performance among other methods. The proposed technique detected
modified plain copy-move and duplicate regions on several datasets, including
CoMoFoD, MICCF220, MICC-F2000, and MICC-F600, in 03.840 s with 91.95%
accuracy. In addition to these findings on detection and accuracy, there is room for
improvement in enhancing detection accuracy to address more intricate and chal-
lenging instances of copy-move forgeries. Moreover, the method should be extended
to encompass various other forms of forgeries (Table 2).

13
 Multimedia Tools and Applications

Table 2  This table summarizes the work related to Key-point or feature-based methods
Paper Title Publication year Method used Performance Perfor-
mance
parameter

“A hybrid copy-move image forgery 2020 SIFT 96.97% F-Score

detection technique based on
Fourier-Mellin and scale invariant
feature
transforms” [60]
“A Robust Copy-Move Forgery 2020 SURF 91.95% Recall
Detection In Digital Image Foren-
sics Using SURF” [61]

b) Block-based methods

Copy-move fraud detection in block-based techniques, the frequency domain charac-

teristics are identified by dividing the altered images into overlapped or non-overlapped
blocks [59].
i. Frequency domain-based methods
Methods based on the frequency domain are used to transform an image in frequency
domain from spatial domain to detect copy-move forgery [62]. Here we have discussed a
few frequency domain methods to detect copy-move fraud. Generally, the following meth-
ods are used for the conversion of an image into a frequency domain:

• Discrete-wavelet transform
• Dyadic-wavelet transform
• Discrete-cosine transform
• Fourier-Mellin transform
• Fast-Fourier transform

Discrete wavelet transformation (DWT) splits the input image into four sub-blocks
called Approximation coefficient block ­(AC), Horizontal coefficient block ­(HC), Vertical
coefficient block (­VC), and Diagonal coefficient block ­(DC) [63]. Figure 12 divides the
input image into four sub-blocks ­(AC, ­HC, ­VC, ­DC ) [63].

• [63] proposed an efficient technique to trace copy-move fraud using Discrete-wavelet-

transform (DWT). Their study found that DWT is much better for identifying image
properties on edges or sudden variations in color contrast than Fourier transform. Dis-
crete-wavelet transform minimizes the image’s size, leading to less time complexity.
The result obtained in this study over the CoMoFoD dataset proved that this proposed
technique could be utilized in copy-move fraud identification and localization with sat-
isfactory performance. As author’s come to the conclusion that the accuracy perfor-
mance of the algorithms is affected by the dimensions of the altered regions.Therefore,
still a need of improving the detection performance and examining the robustness of
the model over other image attacks. Table 3 demonstrates the performance of the intro-
duced method (Table 3).

13
Multimedia Tools and Applications

Table 3  Demonstrates the Method Precision Recall F1-Score

performance of the introduced
method DWT 98.678% 96.387% 97.523%

Fig. 12  Division of the image

into sub-blocks using DWT

• Due to the scale-invariant and translation properties of the Fourier Mellin Transform
(FMT), this method is used to extract features from the image during copy-move fraud
detection [64]. [60] introduced a technique to detect copy-move fraud. To extract fea-
tures from the smooth region of the image, they used Fourier-Mellin transform. In con-
trast, they used the SIFT method to bring out features from the texture region. The
findings of this study are very promising. This proposed method has 96.97% perfor-
mance in terms of F-measure. The suggested solution can be extended for detecting
copy-move frauds in videos.
• Discrete Cosine Transform (DCT) transforms an image into the frequency domain. One
profit of using DCT is that it is invariant to blurring [65]. [44] introduced an algorithm
to detect copy-move fraud in this work, and they used DCT to bring out features of the
photograph. In this study, copy-move fraud detection has the following four steps (See
Fig. 13):

Experimental results of this study proved that the introduced way detected the altered
regions of the image efficiently In the future, it is possible to expand the proposed research
to identify manipulated sections within digital images using error-level analysis and quan-
tization tables as the basis. There is an extreme concern in the proposed solution that the
performance is tested over only 5 images so this issue must be taken into consideration.

• Fast-Fourier-Transform is used to quickly calculate the DFT, which means the matrix cor-
responding to an image is efficiently converted into Fourier coefficients (FFT). [66] Intro-
duced a new method to identify the copy-move fraud in which they used Local features
and Fast Fourier Transform of the altered photograph. The findings of this study demon-
strate that the accuracy of the proposed technique (FFT-ELTP) was 86.62 percent on the
compressed images of the used dataset. Due to the inclusion of intricate transformations
like DCT and FFT in the proposed solution, the overall complexity of the methodology
is heightened. Future research in the same vein may seek to minimize the requirement for

13
 Multimedia Tools and Applications

Fig. 13  Steps of the proposed forgery detection method

such intricate operations. Additionally, investigating the detection of image forgery’s spe-
cific location is also an area to be delved into in further studies.
• [67] Proposed a new approach in which they used Discrete wavelet transform and Dyadic
wavelet transform with Speeded-Up robust features (SURF). The experimental result of
this study shows that Dyadic wavelet transform, when used with SURF, has better recall
and precision but higher execution time than discrete wavelet transforms when used with
SURF. The genuine significance of this suggested research would be fully recognized once
it’s adapted to identify video forgeries. Additionally, the authors are interested in evaluat-
ing how well this CMF detector performs on integrated platforms. Moreover, they aim
to assess the effectiveness of alternative invariant feature transformations like PCA-SIFT
when combined with wavelet transforms. Table 4 compares the two proposed methods of
this study (Table 4).

Table 5 summarizes the different methods and their performance in copy-move fraud
detection.
ii. Local binary pattern

Table 4  This table compares the Introduced Method Recall False Predic- Execution
two proposed methods of this tion rate time in
study seconds

DWT + SURF 64% 19% 2.05

Dyadic WT + SURF 76% 23% 63.122

13
 Multimedia Tools and Applications

Table 5  This table summarizes the techniques working on the frequency domain
Paper Title Publication year Method used Performance Performance parameter

“An Efficient Forensic Approach for Copy-move Forgery Detection via Discrete Wavelet Trans- 2020 DWT 98.678% Precision
form” [63]
“A Robust Copy-Move Forgery Detection In Digital Image Forensics Using SURF” [60] 2020 FMT 96.97% F- score
“Block-based copy–move image forgery detection using DCT"[44] 2019 DCT Efficient In terms of execution time
“Detection of Digital Image Forgery using Fast Fourier Transform and Local Features” [66] 2019 FFT 86.62% Accuracy
“Fast and Robust Copy-Move Forgery Detection Using Wavelet Transforms and SURF” [67] 2016 DyWT 76% Recall

13
 Multimedia Tools and Applications

The local binary pattern (LBP) is the most effective local feature descriptor; it calcu-
lates the texture distribution rule of an image based on the signs of differences between
adjacent pixels to achieve intensity and rotation-invariance [68]. Here we have discussed a
simple Local binary pattern method. [69] discussed the passive forgery detection scheme
using the LBP. LBP is an easy and economic texture operator generally used for texture
classification. LBP is also utilized for counterfeit detection because Image fraud activi-
ties change the pixel values. This LBP label is created corresponding to every pixel of the
residual image. Initially, every pixel which belongs to the residual image is assumed to be
the center pixel then their 8-neighbours are examined. The intensity value of every pixel in
the 8-neighborhood is compared with the value of a central pixel, and then a binary digit
(0,1) is assigned corresponding to the following condition:
{
1, 𝐓𝐚,𝐛 > 𝐂𝐩,𝐪
T a,b =
0, Otherwise (8)

In Eq. (8), ­Cp,q is the pixel in the center of 8-neighbors, and ­Ta,b is the pixel from the
8-neighbors with co-ordinates a and b. An 8-bit binary integer is generated by using the
binary value of each neighboring pixel which is used for the label of the central pixel.
These 8-bit binary numbers can be converted into decimal numbers for the purpose of bet-
ter handling. This process will be used for creating the whole LBP image of the same size.
The study’s findings show that utilizing the BEST-q-CLASS feature selection approach to
compute LBP using noise residuals and co-occurrence matrices leads to a model that per-
forms effectively for practically any set of changes with an accuracy of 98.4%. In the future,
we can enhance model performance by incorporating additional robust features using these
approaches, creating more efficient models with improved performance. Furthermore, as
we identify more types of forgeries, we can apply the scheme outlined to address them
effectively.
iii Dimension reduction-based methods
Dimensionality reduction is the process of transforming data from a high-dimensional
space to one with fewer dimensions so that the conclusions drawn from the reduced dataset
are reasonably close to those drawn from the analysis of the original dataset [70]. Dimen-
sion reduction methods help to increase matching processing speed [71]. Here we have
discussed the two famous dimensionality reduction methods; Principal component analysis
(PCA) and Singular value decomposition (SVD).

• PCA decreases the dimensions of an extensive data set into smaller ones by reducing
the number of variables. Principle component analysis decreases the variables in the
data set while preserving as much information as possible. [71] Discussed new methods
of copy-move fraud detection in which they used kernel PCA with DCT. Tables 6, 7,
and 8 show the experimental result of the proposed study with additive white Gaussian
noise (AWGN), JPEG compression, and Gaussian blurring, respectively. Nonetheless,
the introduction of the segmentation module and dual-branch structure in SD-Net has
increased the method’s complexity. There is a need for future research to explore meth-
ods that can reduce complexity without compromising accuracy. Additionally, thor-
oughly investigating the detection of forgeries in regions that are similar but genuine is
essential.

13
Multimedia Tools and Applications

Table 6  Forgery detection result Performance parameter Kernel size Kernel size
with Gaussian blurring (ω) = 5 Blurring (ω) = 5 Blurring
radius(δ) = 0.5 radius(δ) = 1
24 × 24 40 × 40 24 × 24 40 × 40

Precision 00.9880 00.9980 00.9800 00.9950

Recall 1.0 1.0 00.9830 1.0
Tru positive rate 00.9340 00.9550 00.8590 00.8750
False positive rate 00.0350 00.0180 00.0540 00.0170

Table 7  Forgery detection result Performance parameter SNR = 35 dB SNR = 40 dB

with additive white Gaussian
noise (AWGN) 24 × 24 40 × 40 24 × 24 40 × 40

Precision 00.9790 00.9930 00.9800 1.0

Recall 00.9840 00.9980 00.9910 1.0
Tru positive rate 00.9790 00.9920 00.9850 00.9950
False positive rate 00.0550 00.0460 00.0490 00.0270

Table 8  Forgery detection result Performance parameter Quality factor Q = 80 Quality factor
with JPEG compression Q = 90
24 × 24 40 × 40 24 × 24 40 × 40

Precision 00.9190 00.9330 00.9700 00.9330

Recall 00.9250 00.9380 00.9750 00.9810
Tru positive rate 00.7910 00.9090 00.9570 00.9750
False positive rate 00.0170 00.0130 00.0130 00.0090

In Table 8, SNR stands for signal-to-noise ratio.

In Tables 6, 7 and 8, 24×24, and 40×40 is the forged area size in the image.

• [72] introduced a method of copy-move fraud detection in which they used PCA with
SIFT and DBSCAN. The findings of this study show that the proposed approach suc-
cessfully detects copy-move fraud in the MICC-F220 image data set with an accuracy
of 97%. The suggested approach can be extended to detect other forgeries like splicing
etc.
• [73] suggested a new clone identification method in digital images using Singular value
decomposition (SVD) for data reduction. The SVD method is used to convert corre-
lated variables into non-correlated variables, which can suitably explain the different
relationships between various data items. According to this study, SVD is used for data
reduction where a matrix ­Tcr is defined as:

T cr = Occ × Dcr × 𝐇𝐓𝐫𝐫 (9)

13
 Multimedia Tools and Applications

In equation (9), ­Occ is a matrix of size c × c, which is orthogonal in nature, the size
­ cr matrix is c × r, which is diagonal in nature, and ­HTrris a transpose matrix of a
of the D
matrix ­Hrr of size r × r, which is orthogonal in nature. Findings of the proposed study show
that the DWT-based technique optimizes the search time and accurately locates the forged
region. Whereas DWT followed by the SVD approach further minimized the search time
while preserving the accuracy. Nonetheless, these techniques prove ineffective in identi-
fying alterations in images that have undergone JPEG compression. Regrettably, in the
present day, the majority of images are accessible in JPEG format. One can expand this
research to encompass the detection of tampering in JPEG images as well.

2.2.3 Splicing detection

Image splicing is a forgery algorithm for creating altered image using at least two images
[74–76]. Here we have provided a few methods for splicing detection. Here we have dis-
cussed the machine learning-based splicing detection model and pre-trained residual net-
work-based deep learning CNN techniques-based splicing detection model.

• [77] Proposed a ML based system for detecting splicing in digital images. This scheme
creates a gray-level image from the (Red-Green-Blue) RGB image, and features are
extracted from the image. A feature vector is created by joining these pertinent collec-
tions of characteristics. To train and categorize fake and real images, a machine learn-
ing classifier called logistic regression is utilized. Table 9 shows the correctness of the
proposed method on the different data sets. In the future, the identification of manipu-
lated objects within an image could potentially be achieved using a comparable feature
set in combination with machine learning methods. Additionally, for identifying and
pinpointing copy-move forgeries, utilizing the most relevant feature set in combination
with a co-occurrence texture measure feature could be advantageous.

• [78] Proposed an approach to detect splicing in digital images. In this study, a pre-
trained residual network-based CNN technique has been explored. Feature extrac-
tion has been performed using the RESNET-50 (a pre-trained residual network). This
method trains the classifier model utilizing three classifiers: K-NN, Naive Bayes, and
Multiclass Model using SVM Learner. In the proposed scheme, MATLAB is used to
perform the experiment. Table 10 demonstrates the accuracy of the proposed approach
corresponding to the different classifiers. The performance of the proposed solution
must be examined when replacing the ResNet-50 with its alternative deep-learning
models such as VGGNET and DewnseNet.

Table 9  Shows the accuracy of the proposed work

Paper Title Data set Accuracy achieved

“A technique for image splicing detection using the hybrid feature set” CASIA v1 98.3%
[77] CASIA v2 99.5%
COLUMBIA 98.8%

13
 Multimedia Tools and Applications

Table 10  Demonstrates the proposed model’s accuracy with all three classifiers
Paper title Feature extraction Data set Classifiers Accuracy
achieved

“Image Splicing detection using Deep Residual Network” [78] ResNet-50 CASIA v2 Naïve Bayes 59.91%
(a pre-trained residual network) K-NN 59.91%
Multiclass model using 70.26%
SVM Learner

13
13
Table 11  This table summarizes the techniques based on deep-learning
Paper Title Publication year Method used Performance Perfor-
mance
parameter

“Copy-move forgery detection and localization using a generative adversarial network and 2019 CNNs 95% (approx) Accuracy
convolutional neural-network” [80] and
GANs
“Convolutional Neural Network for Copy-Move forgery detection” [81] 2019 CNN 90% Accuracy
“Dual branch convolutional neural network for copy move forgery detection "[82] 2020 CNN 96% Accuracy
“Digital Image Forgery Detection Using Deep Autoencoder and CNN Features” [79] 2021 Autoencoders and CNN For JPEG 95.9% Accuracy
and for TIFF
93.3%
“CNN-Based Copy-Move Forgery Detection Using Rotation-Invariant Wavelet Feature” [83] 2022 VGG16 CNN 90.5% F1-score
“Image Forgery Detection Using Integrated ConvolutionLSTM (2D) and Convolution (2D)” 2023 ConvLSTM and CNN 85% Accuracy
[84]
Multimedia Tools and Applications
Multimedia Tools and Applications

3 Deep‑learning based Image forgery detection

Apart form the traditional aproches to detect image forgery, recent advances in Artificial
Intelligence (AI) and Machine Learning (ML) allows us to develop new methods which
can successfully identify available forgeries in images. Recently Machine-learning models
like Convolutional Neural Networks (CNNs) and Generative Adversarial Network (GANs),
Support Vector Machine, Auto Encoders, and Convolutional Long Short-Term Memory
(ConvLSTM) have been widely utilized in the area of image forgery detection [27].

• [79] introduced a deep-learning model to identify forgery (copy-move or splicing)

in digital images by keeping in the mind that Image forgery method should be inde-
pendent from the suspected image format. Suggested model utilized multiple structure
stacked autoencoders (SAE) to identify forgery in images. To extract feature from the
image the they used VGG!6 and AlexNet. Finally, for the classification purpose they
used Ensemble Subspace Discriminant classifier. Form the experimnts performed over
the CASIA verion 1.0 and 2.0 it has been observed that the introduced method obtained
95.9% and 93.3% for JPEG and TIFF images respectively. Apart from the accuracy this
approach produces promising results in terms of execution time. This approach only
classify the images as forged or genuine image but not localize the forgery in the image
therefore this work can also be extended for the forgery localization which is an impor-
tant issue in image forgery detection.
• [80] introduced a hybrid model to detect copy-move attack in digital images. Authors
have utilized CNN and Generative Adversarial Network (GAN) for feature extraction
and copy-move symptoms detection respectively. A linear Support Vector Machine
(SVM) is used for the classification purpose. The proposed work has utilized CIFAR-
10, MNIST, MICC-F600, Oxford buildings, and Image Manipulation (IM) data sets for
training and testing purpose. Experiments of the study illustrates that the introduced
scheme is trustworthy in copy-move detection and yield to accuracy of ~ 95%. Authors
have recommended to include more dataset like ImageNet for the training of GAN to
enhance the robustness of the proposed model.
• [81] introduces a purely CNN based model to trace the copy-moved regions in the
suspected image if any. They added a pre-processing layer in their proposed model to
obtain the satisfactory performance. The proposed model architecture contains a pre-
processing layer, three convolutional layer, and two fully connected layer. Rectified Lin-
ear Unit (ReLU) is used as an activation function and Pooling layer uses the MaxPool-
ing and AvgPooling strategy. Finally, SoftMax classifier is used for the classification
purpose. Introduced work has utilized CIFAR-10, MNIST, MICC-F600, Oxford build-
ings, and Image Manipulation (IM) data sets for training and testing purpose. Experi-
ments of the study illustrates that the introduced scheme is trustworthy in copy-move
detection and yield to accuracy of 90% with a set iteration limit. Authors suggested
that other network architectures could be investigated in related future research, and
in-depth studies could be carried out by utilizing a larger dataset than the one employed
here.
• [82] implemented an idea of dual branch CNN to detect copy-move forgery. They have
used two different branches of CNN for feature extraction and forgery detection in the
suggested model. This strategy is tested over the various kernel size combinations.
Experimental results over the MICC F2000 data set illustrates that the suggested model
is trustworthy in copy-move detection and yield to accuracy of 96%. Authors of the

13
 Multimedia Tools and Applications

study suggested that the proposed model can be examined and evaluated over datasets
of different image sizes. This model can also be utilized in the detection of other for-
gery operations.
• [83] given a rotation invariant feature-based copy-move detection scheme using a pre-
trained CNN model VGG16. The proposed technique consists of four basic modules:
rotation invariant, extraction of feature, correlation, and mask decoder module. CoMo-
FoD, MICC-F2000, D, and COVERAGE data sets for training and testing purpose.
Experiments of the study illustrates that the introduced scheme is trustworthy in copy-
move detection and yield to F1 score of 90.5%. Suggested solution can be extended to
detect other image forgeries like splicing etc.
• [84] developed a model for image forgery detection. Suggested model classify the
suspected image as genuine or fake classes. For this purpose they have proposed
two different ways of forgery detection. In the first way, ConvLSTM(1D) is uti-
lized with Convolutional(2D). In the second way, ConvLSTM(2D) is utilized with
Convolutional(2D). Both the approaches tested over the CASIA v.2.0 data set and it
is found that second approach is efficient than first one in terms of accuracy as first
approach results in 72% accuracy and second resultrs in 85% accuracy. Suggested
approach must be extended for improving the accuracy, apart from this, suggested solu-
tion can be examined over various image alternation technioques (Table 11).

4 Image forensic data sets

In Digital image forensics, a dataset is a carefully managed collection of digital images that
a researcher uses to train, test, and assess the effectiveness of the proposed algorithms for
forgery detection [85]. It is observed that the same methods of forgery detection perform
differently with different data sets. As researchers, we must develop methods whose per-
formance does not depend on the data set. We have provided a few commonly used data
sets to assess the performance of the proposed forgery detection methods (See Table 12).
Table 12 has four columns, the first column (Name of image dataset) shows the name of
the dataset with the citation where this dataset has been used, second column (Dataset
used for) shows the purpose of utilizing dataset in mentioned citation, third column (Per-
formance) shows the achieved performance when the dataset was used, and fourth column
(Dataset description) describes the characteristic of the dataset in short.
Here we have provided the links to download publicly available datasets:

CASIA v1.0 and CASIA v2.0: https://​www.​kaggle.​com/​datas​ets/​sopha​tvath​ana/​casia-​

datas​et
COLUMBIA: https://​www.​ee.​colum​bia.​edu/​ln/​dvmm/​downl​oads/​AuthS​plice​dData​Set/​
AuthS​plice​dData​Set.​htm
CoMoFoD: https://​www.​vcl.​fer.​hr/​comof​od/​downl​oad.​html
MICC-F220, MICC-F2000, MICC-F8multi, and MICC-F600: http://​lci.​micc.​unifi.​
it/​labd/​2015/​01/​copy-​move-​forge​ry-​detec​tion-​and-​local​izati​on/
GRIP: https://​figsh​are.​com/​colle​ctions/​Source-​Desti​nation_​Foren​sics_​CMFD_​Datas​et/​
62211​92
CIFAR-10: https://​www.​cs.​toron​to.​edu/​~kriz/​cifar.​html
Copy-move hard: http://​dx.​doi.​org/https://​doi.​org/​10.​6084/​m9.​figsh​are.​978736
COVERAGE: https://​github.​com/​wenbi​han/​cover​age

13
 Table 12  Comparison of various image data set
Name of the Dataset used for Performance Dataset description
image dataset

CASIA v1.0 [77] Splicing detection Accuracy = 98.3% This data set has two image directories, Au and Sp, with 800 and 921 files, respec-
tively [86]
CASIA v2.0 Splicing detection Accuracy = 99.5% This data set has three image directories, Au, Tp, and CASIA 2 Groundtruth, with
[77] 7492, 5125, and 5123 files, respectively [87]
COLUMBIA Splicing detection Accuracy = 98.8% The data set includes 1845 image blocks (128 × 128), of which 933 are authentic,
[77] and 912 are spliced image blocks [88]
Multimedia Tools and Applications

CoMoFoD Copy-move Precision = 96.845% CoMoFoD database contains 260 manipulated or altered photos, divided into two
[89] detection Recall = 92.5% groups (small 512 × 512 and large 3000 × 2000) [90, 91]
F1-score = 94.35%
MICC-F220 Copy-move False positive rate = 3.1% Within this dataset, there are 220 images, with an equal split of 110 forged images
[92] detection Recall = 99.2% and 110 authentic images [93]
MICC-F2000 Copy-move False positive rate = 6.8% This dataset comprises 2,000 images, with 700 of them being forgeries and 1,300
[92] detection Recall = 98.5% being genuine/authentic images [93]
MICC-F8multi Copy-move False positive rate = 6.9% This data set has eight altered images with realistic multiple cloning[94]
[92] detection Recall = 98.8%
MICC-F600 Copy-move Recall = 100% This data contains 440 authentic images, of which 160 are altered images, and 160
[93] detection True negative rate = 100% are ground truth images [93]
GRIP Copy-move Precision = (91.76%) 80 legitimate and 80 altered images with a size of 768 × 1024 are used in the GRIP
[95] dataset [95]
Copy-move hard (CMH) Copy-move The average accuracy on CMH data set has 108 images. Out of 108 images, 23 images are copy moved
[96] ­(CMHp1) is = ­(CMHp1), 25 Images rotate the copied region ­(CMHp2), 26 images with resizing
96.19% of the cloned area ­(CMHp3), and 34 images use rotation and scaling ­(CMHp4)
[97]
Dresden Image Database [54] Imaging device identificaT Accuray = 100% In total, there are over 14,000 images available, encompassing various camera
configurations, environmental conditions, and specific scenes [98]
CIFAR-10 [80] Copy-move Accuracy = 95% (approx) 60,000 color images have been used in the CIFAR-10 dataset, each sized at 32 × 32
pixels, distributed across ten categories. 6000 images has been kept in each
category. The dataset is further separated into 50,000 and 10,000 images for
training and testing purpose respectively [99]

13
 Table 12  (continued)
Name of the Dataset used for Performance Dataset description
image dataset

13
D data set [83] Copy-move F1-score = 90.5% It consists of images of moderate size, with most being either 700 × 1000 or
1000 × 700 in dimensions, and it is additionally segmented into multiple datasets
labeled as D0, D1, and D2. [100]
COVERAGE [83] Copy-move F1-score = 90.5% A novel database has been created, which includes manipulated images where
objects have been copied and moved, along with their authentic counterparts
featuring similar but real objects. This database, known as COVERAGE, aims
to emphasize and tackle the problem of uncertainty in tamper detection methods
caused by the inherent similarity found in unaltered natural images [101]
Image Manipulation dataset [80] Copy-move Accuracy = 95% (approx) The Image Manipulation Dataset serves as an authentic reference collection used
to assess the accuracy of identifying alterations in images, such as tampering
artifacts. It contains 48 primary images, distinct sections extracted from these
images, and a software framework for producing accurate reference data[102]
MNIST dataset [81] Copy-move Accuracy = 90% The dataset includes a training batch with 60,000 images and a testing batch with
10,000 images, all depicting handwritten digits. These images of digits have been
standardized in size and positioned at the center, each occupying a consistent
area of 28 × 28 pixels [103]
Multimedia Tools and Applications
Multimedia Tools and Applications

Image Manipulation Dataset (IMD): https://​www5.​cs.​fau.​de/​resea​rch/​data/​image-​

manip​ulati​on/
Dresden image dataset: https://​www.​kaggle.​com/​datas​ets/​hanju​nyang1/​dresd​en
D dataset: http://​www.​dicgim.​unipa.​it/​cvip/ or by personal request to author.
MNIST dataset: https://​figsh​are.​com/​artic​les/​datas​et/​data/​14132​507

5 Conclusions

In this study, we have tried to provide a complete summary of various techniques used in
blind forgery detection, especially Imaging device identification,copy-move detection, and
splicing detection. We have been attempting to explore PRNU, CRF, and feature extrac-
tion-based methods to identify imaging devices. While for copy-move fraud detection, we
have focused on the Key-point and Block based methods. In this survey paper, we have
provided information about the data set in tabular form. To better understand this study for
any reader, we have tried to summarize the survey by using various figures and tables. To
preserve the importance of this study for future researchers, we have used the most recent
and up-to-date references. In conclusion, we aspire that this study will prove beneficial to
researchers within the digital forensics field, specifically those concentrating on blind or
passive forgery detection.
Acknowledgements I am (Deependra Kumar Shukla) grateful to the UGC and the Government of India for
granting me the UGC- (JRF/SRF) fellowship, which enables me to pursue my research endeavors.

Author Contributions The authors contributed equally to this work.

Funding No funding was received to assist with the preparation of this manuscript.

Data Availability No additional data or material has been used for this work other than the referenced
papers.

Code Availability No code has been developed by the authors for this work.

Declarations
Conflict of Interests The authors declare that they have no conflict of interest.

References
1. Jana M, Jana B, Joardar S (2022) Local feature based self-embedding fragile watermarking scheme
for tampered detection and recovery utilizing AMBTC with fuzzy logic, J King Saud Univ Comput
Inf Sci, no. xxxx, 2021, https://​doi.​org/​10.​1016/j.​jksuci.​2021.​12.​011
2. Raju PM, Nair MS (2018) Copy-move forgery detection using binary discriminant features. J King
Saud Univ Comput Inf Sci 34(2):165–178. https://​doi.​org/​10.​1016/j.​jksuci.​2018.​11.​004
3. Sekhar PC, Shankar TN (2023) An object-based splicing forgery detection using multiple noise fea-
tures. Multimed Tools Appl. https://​doi.​org/​10.​1007/​s11042-​023-​16534-z
4. Verma M, Singh D (2023) Survey on image copy-move forgery detection. Multimed Tools Appl.
https://​doi.​org/​10.​1007/​s11042-​023-​16455-x
5. Sushir RD, Wakde DG, Bhutada SS (2023) Enhanced blind image forgery detection using an accu-
rate deep learning based hybrid DCCAE and ADFC. Multimed Tools Appl. https://​doi.​org/​10.​1007/​
s11042-​023-​15475-x

13
 Multimedia Tools and Applications

6. Abir NAM, Warif NBA, Zainal N (2023) An automatic enhanced filters with frequency-based copy-
move forgery detection for social media images. Multimed Tools Appl. https://​doi.​org/​10.​1007/​
s11042-​023-​15506-7
7. Li Q, Wang C, Zhou X, Qin Z (2022) Image copy-move forgery detection and localization
based on super-BPD segmentation and DCNN. Sci Rep 12(1):14987. https://​doi.​org/​10.​1038/​
s41598-​022-​19325-y
8. Ferreira WD, Ferreira CBR, da Cruz Júnior G, Soares F (2020) A review of digital image forensics,
Comput Electr Eng, vol. 85 https://​doi.​org/​10.​1016/j.​compe​leceng.​2020.​106685
9. Dhanaraj RS, Sridevi M (2021) A study on detection of copy-move forgery in digital images, in Pro-
ceedings of the 3rd International Conference on Intelligent Communication Technologies and Virtual
Mobile Networks, ICICV 2021, pp. 900–905. https://​doi.​org/​10.​1109/​ICICV​50876.​2021.​93885​76
10. Uma S, Sathya PD (2019) A detailed review of copy-move forgery detection in digital image. Glob J
Eng Sci Res. https://​doi.​org/​10.​5281/​zenodo.​25378​23
11. Ansari MD, Ghrera SP, Tyagi V (Jan.2014) Pixel-based image forgery detection: A review. IETE J
Educ 55(1):40–46. https://​doi.​org/​10.​1080/​09747​338.​2014.​921415
12. What is Photo Retouching? Why It’s So Important to Retouch. https://​www.​imagi​nated.​com/​photo​
graphy/​photo​graphy-​gloss​ary/​what-​is-​photo-​retou​ching/ (accessed Sep. 20, 2022)
13. AlZahir S, Hammad R (2020) Image forgery detection using image similarity. Multimed Tools Appl
79(39–40):28643–28659. https://​doi.​org/​10.​1007/​s11042-​020-​09502-4
14. Rajput A (2018) Image Splicing | Set 1 (Introduction) - GeeksforGeeks. https://​www.​geeks​forge​eks.​
org/​image-​splic​ing-​set-1-​intro​ducti​on/ (accessed Sep. 20, 2022)
15. Koul S, Kumar M, Khurana SS, Mushtaq F, Kumar K (2022) An efficient approach for copy-move
image forgery detection using convolution neural network. Multimed Tools Appl 81(8):11259–11277.
https://​doi.​org/​10.​1007/​s11042-​022-​11974-5
16. Meena KB, Tyagi V (2023) Image splicing forgery detection using noise level estimation. Multimed
Tools Appl 82(9):13181–13198. https://​doi.​org/​10.​1007/​s11042-​021-​11483-x
17. Kaur N, Jindal N, Singh K (2020) A passive approach for the detection of splicing forgery in digital
images. Multimed Tools Appl 79(43–44):32037–32063. https://​doi.​org/​10.​1007/​s11042-​020-​09275-w
18. Kaur A, Rani J (2016) Digital Image Forgery and Techniques of Forgery Detection: A brief review.
International Journal of Technical Research & Science 1(4):18–24
19. Raja K, Gupta G, Venkatesh S, Ramachandra R, Busch C (2022) Towards generalized morphing
attack detection by learning residuals. Image Vis Comput 126:104535. https://​doi.​org/​10.​1016/j.​ima-
vis.​2022.​104535
20. Image Processing : Morphing (1997) https://​www.​owlnet.​rice.​edu/​~elec5​39/​Proje​cts97/​morph​jrks/​
morph.​html (accessed Sep. 20, 2022)
21. Thakur T, Singh K, Yadav A (2018) Blind Approach for Digital Image Forgery Detection. Int J Com-
put Appl 179(10):34–42. https://​doi.​org/​10.​5120/​ijca2​01891​6108
22. Hegazi A, Taha A, Selim MM (2021) An improved copy-move forgery detection based on density-
based clustering and guaranteed outlier removal. J King Saud Univ Comput Inf Sci 33(9):1055–1063.
https://​doi.​org/​10.​1016/j.​jksuci.​2019.​07.​007
23. Vijayalakshmi NVSK, Sasikala KJ, Shanmuganathan C (2023) Copy-paste forgery detection
using deep learning with error level analysis, Multimed Tools Appl, https://​doi.​org/​10.​1007/​
s11042-​023-​15594-5
24. Yang B, Li Z, Zhang T (2020) A real-time image forensics scheme based on multi-domain learning. J
Real-Time Image Process 17(1):29–40. https://​doi.​org/​10.​1007/​s11554-​019-​00893-8
25. Liu K et al (2019) Copy move forgery detection based on keypoint and patch match. Multimed Tools
Appl 78(22):31387–31413. https://​doi.​org/​10.​1007/​s11042-​019-​07930-5
26. Liu G, Reda FA, Shih KJ, Wang TC, Tao A, Catanzaro B (2018) Image inpainting for irregular holes
using partial convolutions. In: Proceedings of the European conference on computer vision (ECCV),
pp 85–100
27. Zanardelli M, Guerrini F, Leonardi R, Adami N (2023) Image forgery detection: a survey of recent
deep-learning approaches. Multimed Tools Appl 82(12):17521–17566. https://​doi.​org/​10.​1007/​
s11042-​022-​13797-w
28. He L, Qiang Z, Shao X, Lin H, Wang M, Dai F (2022) Research on High-Resolution Face Image
Inpainting Method Based on StyleGAN. Electron 11(10):1–18. https://​doi.​org/​10.​3390/​elect​ronic​
s1110​1620
29. Qiao T, Zhu A, Retraint F (2018) Exposing image resampling forgery by using linear parametric
model. Multimed Tools Appl 77(2):1501–1523. https://​doi.​org/​10.​1007/​s11042-​016-​4314-1
30. Alamro L, Yusoff N (2017) Copy-move forgery detection using integrated DWT and SURF. J Tel-
ecommun Electron Comput Eng 9(1–2):67–71

13
Multimedia Tools and Applications

31. Sharma P, Kumar M, Sharma H (2022) Comprehensive analyses of image forgery detection
methods from traditional to deep learning approaches: an evaluation. Multimed Tools Appl
82(12):18117–18150. https://​doi.​org/​10.​1007/​s11042-​022-​13808-w
32. Koundinya Anjan K, Sunanda D, Mahesh G, Sneha S (2022) Characteristic overview of digital
image forensics tools. In: Proceedings of the 2nd International Conference on Recent Trends in
Machine Learning, IoT, Smart Cities and Applications: ICMISC 2021. Springer, pp 157–162
33. Hosny KM, Mortda AM, Fouda MM, Lashin NA (2022) An efficient cnn model to detect copy-
move image forgery. IEEE Access 10:48622–48632. https://​doi.​org/​10.​1109/​ACCESS.​2022.​31722​
73
34. Fadhil JM, Trupti B (2022) An efficient technique for image forgery detection using local binary
pattern (hessian and center symmetric) and transformation method. Scientific Journal Al-Imam
University College 1:1–11
35. Manna N, Kumar S, Kakar R, Nayak S, Rout JK, Kumar Balabantaray B (2022) IFChatbot: Con-
volutional Neural Network based chatbot for Image Forgery Detection and Localization, in 2022
IEEE India Council International Subsections Conference (INDISCON), pp. 1–6. https://​doi.​org/​
10.​1109/​INDIS​CON54​605.​2022.​98629​26
36. Alhaidery MMA, Taherinia AH (2022) A passive image forensic scheme based on an adap-
tive and hybrid techniques. Multimed Tools Appl 81(9):12681–12699. https://​doi.​org/​10.​1007/​
s11042-​022-​12374-5
37. Kadam K, Ahirrao S, Kotecha K (2021) AHP validated literature review of forgery type dependent
passive image forgery detection with explainable AI. Int J Electr Comput Eng 11(5):4489–4501.
https://​doi.​org/​10.​11591/​ijece.​v11i5.​pp4489-​4501
38. Sai Achyuth P, Satyanarayana V (2021) Image forgery detection techniques: a brief review. In:
Proceedings of Second International Conference in Mechanical and Energy Technology: ICMET
2021, India. Springer, pp 351–357
39. Subramanian N, Elharrouss O, Al-Maadeed S, Bouridane A (2021) Image Steganography: A
Review of the Recent Advances. IEEE Access 9:23409–23423. https://​doi.​org/​10.​1109/​ACCESS.​
2021.​30539​98
40. Bansal A, Kumar V (2021) Steganography Technique Inspired by Rook, https://​servi​ces.​igi-​
global.​com/​resol​vedoi/​resol​ve.​aspx?​doi=​10.​4018/​IJISP.​20210​40103, vol. 15, no. 2, pp. 53–67,
https://​doi.​org/​10.​4018/​IJISP.​20210​40103
41. Bansal A, Muttoo SK, Kumar V (2016) Security against Sample Pair Steganalysis in Eight Queens
Data Hiding Technique. Int J Comput Netw Inf Secur 8(8):39–46. https://​doi.​org/​10.​5815/​ijcnis.​
2016.​08.​05
42. Begum M, Uddin MS (2020) Digital image watermarking techniques: A review, Information
(Switzerland), vol. 11, no. 2. MDPI AG. https://​doi.​org/​10.​3390/​info1​10201​10
43. Ray A, Roy S (2020) Recent trends in image watermarking techniques for copyright protection: a
survey. Int J Multimed Inf Retr 9(4):249–270. https://​doi.​org/​10.​1007/​s13735-​020-​00197-9
44. Parveen A, Khan ZH, Ahmad SN (2019) Block-based copy–move image forgery detection using
DCT. Iran J Comput Sci 2(2):89–99. https://​doi.​org/​10.​1007/​s42044-​019-​00029-y
45. Meena KB, Tyagi V (2021) Efficient Passive Forgery Detection in Digital Images, Jaypee Univer-
sity of Engineering and Technology, Guna, [Online]. Available: http://​hdl.​handle.​net/​10603/​338230.
Accessed 25/09/2023
46. Liu Y, Zou Z, Yang Y, Law NFB, Bharath AA (2021) Efficient source camera identification with
diversity-enhanced patch selection and deep residual prediction. Sensors 21(14):1–22. https://​doi.​
org/​10.​3390/​s2114​4701
47. Wang B, Wang Y, Hou J, Li Y, Guo Y (2022) Open-Set source camera identification based on
envelope of data clustering optimization (EDCO). Comput Secur, vol. 113 https://​doi.​org/​10.​
1016/j.​cose.​2021.​102571
48. Shukla DK, Bansal A, Singh P (2022) Performance analysis of various copy-move forgery detec-
tion methods. i-Manager’s Journal on Digital Signal Processing 10(2):1
49. Tahaoglu G, Ulutas G, Ustubioglu B, Nabiyev VV (2021) Improved copy move forgery detec-
tion method via L*a*b* color space and enhanced localization technique. Multimed Tools Appl
80(15):23419–23456. https://​doi.​org/​10.​1007/​s11042-​020-​10241-9
50. Wei H, Kehtarnavaz N (2019) Semi-Supervised Faster RCNN-Based Person Detection and Load
Classification for Far Field Video Surveillance. Mach Learn Knowl Extr 1(3):756–767. https://​doi.​
org/​10.​3390/​make1​030044
51. Obeidat AA (2017) Hybrid approach for botnet detection using k-means and k-medoids with Hop-
field neural network. Int J Commun Networks Inf Secur 9(3):305–313

13
 Multimedia Tools and Applications

52. Alhaidery MMA, Taherinia AH, Yazdi HS (2022) Cloning detection scheme based on linear and
curvature scale space with new false positive removal filters. Multimed Tools Appl 81(6):8745–
8766. https://​doi.​org/​10.​1007/​s11042-​022-​12237-z
53. Fanfani M, Piva A, Colombo C (2022) PRNU registration under scale and rotation transform based
on convolutional neural networks. Pattern Recognit 124:108413. https://​doi.​org/​10.​1016/j.​patcog.​
2021.​108413
54. Behare MS, Bhalchandra AS, Kumar R (2019) Source Camera Identification using Photo Response
Noise Uniformity, in Proceedings of the 3rd International Conference on Electronics and Communi-
cation and Aerospace Technology, ICECA 2019, pp. 731–734. https://​doi.​org/​10.​1109/​ICECA.​2019.​
88222​12
55. Flor E, Aygun R, Mercan S, Akkaya K (2021) PRNU-based Source Camera Identification for Multi-
media Forensics, Proc. - 2021 IEEE 22nd Int. Conf. Inf. Reuse Integr. Data Sci. IRI 2021, pp. 168–
175, https://​doi.​org/​10.​1109/​IRI51​335.​2021.​00029
56. Xu B, Wang X, Zhou X, Xi J, Wang S (2016) Source camera identification from image texture fea-
tures. Neurocomputing 207:131–140. https://​doi.​org/​10.​1016/j.​neucom.​2016.​05.​012
57. Grossberg MD, Nayar SK (2003) Determining the camera response from images: What is knowable?,
IEEE Trans Pattern Anal Mach Intell, vol. 25, no. 11, https://​doi.​org/​10.​1109/​TPAMI.​2003.​12401​19
58. Chen C, McCloskey S, Yu J (2019) Analyzing modern camera response functions, in Proceedings
- 2019 IEEE Winter Conference on Applications of Computer Vision, WACV 2019, Mar, pp. 1961–
1969. https://​doi.​org/​10.​1109/​WACV.​2019.​00213
59. Sadeghi S, Dadkhah S, Jalab HA, Mazzola G, Uliyan D (2018) State of the art in passive digital
image forgery detection: copy-move image forgery. Pattern Anal Appl 21(2):291–306. https://​doi.​org/​
10.​1007/​s10044-​017-​0678-8
60. Meena KB, Tyagi V (2020) A hybrid copy-move image forgery detection technique based on Fourier-
Mellin and scale invariant feature transforms. Multimed Tools Appl 79(11–12):8197–8212. https://​
doi.​org/​10.​1007/​s11042-​019-​08343-0
61. Badr A, Youssif A, Wafi M (2020) A robust copy-move forgery detection in digital image foren-
sics using SURF. In: 2020 8th International Symposium on Digital Forensics and Security (ISDFS).
IEEE, pp 1–6
62. Introduction to Frequency domain (2022) https://​www.​tutor​ialsp​oint.​com/​dip/​intro​ducti​on_​to_​frequ​
ency_​domain.​htm (accessed Sep. 19, 2022)
63. Ashraf R et al. (2020) An Efficient Forensic Approach for Copy-move Forgery Detection via Discrete
Wavelet Transform,” 1st Annu Int Conf Cyber Warf Secur ICCWS 2020 - Proc, https://​doi.​org/​10.​
1109/​ICCWS​48432.​2020.​92923​72
64. Pourkashani A, Shahbahrami A, Akoushideh A (2021) Copy-move forgery detection using convolu-
tional neural network and K-mean clustering. Int J Electr Comput Eng 11(3):2604–2612. https://​doi.​
org/​10.​11591/​ijece.​v11i3.​pp2604-​2612
65. Jaiswal AK, Gupta D, Srivastava R (2020) Detection of copy-move forgery using hybrid approach
of DCT and BRISK. In: 2020 7th International Conference on Signal Processing and Integrated Net-
works (SPIN). IEEE, pp 471–476
66. Kanwal N, Girdhar A, Kaur L, Bhullar JS (2019) Detection of digital image forgery using fast fou-
rier transform and local features. In: 2019 international conference on automation, computational and
technology management (ICACTM). IEEE, pp 262–267
67. Hashmi MF, Keskar AG (2019) Fast and robust copy-move forgery detection using wavelet trans-
forms and SURF. Int Arab J Inf Technol 16(2):304–311
68. Luo Q, Su J, Yang C, Silven O, Liu L (2022) Scale-selective and noise-robust extended local binary
pattern for texture classification. Pattern Recognit 132:108901. https://​doi.​org/​10.​1016/J.​PATCOG.​
2022.​108901
69. Farooq S, Yousaf MH, Hussain F (2017) A generic passive image forgery detection scheme using
local binary pattern with rich models. Comput Electr Eng 62:459–472. https://​doi.​org/​10.​1016/j.​
compe​leceng.​2017.​05.​008
70. Nsang AS, Bello AM, Shamsudeen H (2015) Image reduction using assorted dimensionality reduc-
tion techniques. CEUR Workshop Proc 1353(June):139–146
71. Chen H, Yang X, Lyu Y (2020) Copy-move forgery detection based on keypoint clustering and simi-
lar neighborhood search algorithm. IEEE Access 8:36863–36875. https://​doi.​org/​10.​1109/​ACCESS.​
2020.​29748​04
72. Mursi MFM, Salama MM, Habeb MH (2017) An Improved SIFT-PCA-Based Copy-Move Image
Forgery Detection Method. Int J Adv Res Comput Sci Electron Eng 6(3):23–28
73. Mishra M, Chandra Adhikary M, Adhikary FMLt C (2014) Detection of Clones in Digital Images
Digital Image Forgery Detection View project MAKE-meteorological analyser & knowledge

13
Multimedia Tools and Applications

extractor View project Detection of Clones in Digital Images. [Online]. Available: https://​www.​resea​
rchga​te.​net/​publi​cation/​26427​6516. Accessed 15/07/22
74. Jain I, Goel N (2021) Advancements in image splicing and copy-move forgery detection tech-
niques: A survey, Proc Conflu 2021 11th Int Conf Cloud Comput Data Sci Eng, pp. 470–475,
https://​doi.​org/​10.​1109/​Confl​uence​51648.​2021.​93771​04
75. Rao Y, Ni J, Zhao H (2020) Deep Learning Local Descriptor for Image Splicing Detection and
Localization. IEEE Access 8:25611–25625. https://​doi.​org/​10.​1109/​ACCESS.​2020.​29707​35
76. Ahmed B, Gulliver TA, S. alZahir (2020) Image splicing detection using mask-RCNN. Signal,
Image Video Process 14(5):1035–1042. https://​doi.​org/​10.​1007/​s11760-​020-​01636-0
77. Jaiswal AK, Srivastava R (2020) A technique for image splicing detection using hybrid feature set.
Multimed Tools Appl 79(17–18):11837–11860. https://​doi.​org/​10.​1007/​s11042-​019-​08480-6
78. Jaiswal AK, Srivastava R (2019) Image Splicing Detection using Deep Residual Network. SSRN
Electron J. https://​doi.​org/​10.​2139/​ssrn.​33510​72
79. Bibi S, Abbasi A, Haq IU, Baik SW, Ullah A (2021) Digital Image Forgery Detection Using Deep
Autoencoder and CNN Features, Human-centric Comput Inf Sci, vol. 11, https://​doi.​org/​10.​22967/​
HCIS.​2021.​11.​032
80. Abdalla Y, Tariq Iqbal M, Shehata M (2019) Copy-move forgery detection and localization using
a generative adversarial network and convolutional neural-network, Inf, vol. 10, no. 9, https://​doi.​
org/​10.​3390/​info1​00902​86
81. Abdalla Y, Iqbal MT, Shehata M (2019) Convolutional neural network for copy-move forgery
detection. Symmetry 11(10):1280
82. Goel N, Kaur S, Bala R (2021) Dual branch convolutional neural network for copy move forgery
detection, no. December 2020, pp. 656–665, https://​doi.​org/​10.​1049/​ipr2.​12051
83. Lee SI, Park JY, Eom IK (2022) CNN-Based Copy-Move Forgery Detection Using Rotation-
Invariant Wavelet Feature. IEEE Access 10(October):106217–106229. https://​doi.​org/​10.​1109/​
ACCESS.​2022.​32120​69
84. Yogita S, Prashant S, Rawat CSD (2023) Image forgery detection using integrated convolution-
LSTM (2D) and convolution (2D). International Journal of Electrical and Electronics Research
(IJEER) 11(2):631–638
85. Maleve N (2019) An Introduction to Image Datasets | u n t h i n k i n g . p h o t o g r a p h y. https://​
unthi​nking.​photo​graphy/​artic​les/​an-​intro​ducti​on-​to-​image-​datas​ets (accessed Sep. 20, 2022)
86. Sovathana P (2018) Casia dataset | Kaggle. https://​www.​kaggle.​com/​datas​ets/​sopha​tvath​ana/​casia-​
datas​et (accessed Sep. 02, 2022)
87. Goel D (2020) CASIA 2.0 Image Tampering Detection Dataset | Kaggle. https://​www.​kaggle.​com/​
datas​ets/​divg07/​casia-​20-​image-​tampe​r ing-​detec​tion-​datas​et (accessed Sep. 02, 2022)
88. Ng T-T, Chang S-F, Sun Q (2004) A data set of authentic and spliced image blocks. In: ADVENT
Technical Report, vol 4. Columbia University
89. Niyishaka P, Bhagvati C (2020) Copy-move forgery detection using image blobs and BRISK fea-
ture. Multimed Tools Appl. https://​doi.​org/​10.​1007/​s11042-​020-​09225-6
90. Tralic D, Zupancic I, Grgic S, Grgic M (2013) CoMoFoD—New database for copy-move forgery
detection. In: Proceedings ELMAR-2013. IEEE, pp 49–54
91. CoMoFoD (2013) https://​www.​vcl.​fer.​hr/​comof​od/ (accessed Sep. 02, 2022)
92. Soni B, Das PK, Thounaojam DM (2018) multiCMFD: fast and efficient system for multiple copy-
move forgeries detection in image. In: Proceedings of the 2018 international conference on image
and graphics processing, pp 53–58
93. Elaskily MA et al (2020) A novel deep learning framework for copy-moveforgery detec-
tion in images. Multimed Tools Appl 79(27–28):19167–19192. https://​doi.​org/​10.​1007/​
s11042-​020-​08751-7
94. Sadeghi S, Jalab HA, Wong K, Uliyan D, Dadkhah S (2017) Keypoint based authentication and
localization of copy-move forgery in digital image. Malaysian J Comput Sci 30(2):117–133
95. Wang C, Zhang Z, Zhou X (2018) An image copy-move forgery detection scheme based on
A-KAZE and SURF features. Symmetry (Basel) 10(12):1–20. https://​doi.​org/​10.​3390/​sym10​
120706
96. Silva E, Carvalho T, Ferreira A, Rocha A (2015) Going deeper into copy-move forgery detection:
Exploring image telltales via multi-scale analysis and voting processes. J Vis Commun Image
Represent 29:16–32. https://​doi.​org/​10.​1016/j.​jvcir.​2015.​01.​016
97. Al-Qershi OM, Khoo BE (2018) Evaluation of copy-move forgery detection: datasets and evaluation
metrics. Multimed Tools Appl 77(24):31807–31833. https://​doi.​org/​10.​1007/​s11042-​018-​6201-4
98. Gloe T, Böhme R (2010) The dresden image database for benchmarking digital image forensics. J
Digit Forensic Pract 3(2–4):150–159. https://​doi.​org/​10.​1080/​15567​281.​2010.​531500

13
 Multimedia Tools and Applications

99. CIFAR-10 and CIFAR-100 datasets (n.d.) https://​www.​cs.​toron​to.​edu/​~kriz/​cifar.​html (accessed

Sep. 19, 2023)
100. Ardizzone E, Bruno A, Mazzola G (2015) Copy-move forgery detection by matching triangles of
keypoints, IEEE Trans Inf Forensics Secur, vol. 10, https://​doi.​org/​10.​1109/​TIFS.​2015.​24457​42
101. Wen B, Zhu Y, Subramanian R, Ng TT, Shen X, Winkler S (2016) Coverage – a novel database for
copy-move forgery detection. In: 2016 IEEE International Conference on Image Processing (ICIP),
pp 161–165. https://​doi.​org/​10.​1109/​ICIP.​2016.​75323​39
102. Image Manipulation Dataset (n.d.) https://​www5.​cs.​fau.​de/​resea​rch/​data/​image-​manip​ulati​on/
(accessed Sep. 19, 2023)
103. MNIST - Machine Learning Datasets (n.d.) https://​datas​ets.​activ​eloop.​ai/​docs/​ml/​datas​ets/​mnist/
(accessed Sep. 25, 2023)

Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.

Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under
a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted
manuscript version of this article is solely governed by the terms of such publishing agreement and applicable
law.

13