International Journal of Computer Science and Information Security (IJCSIS),

Vol. 14, No. 11, November 2016

Review on Image Splicing Forgery Detection

PNRL Chandra Sekhar Dr. T N Sankar
Research Scholar Professor
Dept of CSE Dept of CSE
KL University, Vijayawada KL University, Vijaywada

Abstract

With the growing usage of internet in daily life along with the
usage of powerful image editing software tools in creating forged
images effortlessly, making us to lose the trust in authenticity of
the images. More than a decade an extensive research is going
on in the field of Image forensics aims at restoring
trustworthiness in images by bringing various tampering
detection techniques. In this regard, we attempt to survey various
techniques found particularly in Splicing Image Forgery
Detection. We summarize both feature based as well as camera
characteristics based techniques over the recent years.

Keywords – Image splicing, Copy-Move Forgery, Re-sampling,

Passive techniques

I. Introduction
Today’s world is living in the remarkable era of visual Fig1[2]: Example of image forgery John Forbes Kerry with
imagery which made it possible to access, process and share Jane Fonda
information very easily. Historically we had confidence in
the integrity of this imagery however; the rapid growth of This manipulation of images is going on from the past and
technological advancement in digital technology in terms of even accepted in areas like forensic investigation,
powerful algorithms, tools such as Photoshop, CorelDraw Information Technology, medical Imaging, Journalism,
for manipulating digital images brought with major security Intelligence service etc. [2] Nowadays organizations
challenges that rise question in this trust. Name a few interested in paperless work and e-government services
particularly from magazines to fashion industry in terms of resulting a huge amount of data stored in digital format and
media outlets, scientific magazines, political campaigns, this gives rise to many challenges to secure authentic data.
courtrooms, photo hoaxes that reached our inbox, doctored Unfortunately, the various collections of data like
photographs etc., are appearing with a growing frequency documents, files, voice data, and image data are all
and sophistication. Then it becomes very difficult to vulnerable to manipulation and doctoring. This gives rises
discriminate which is authentic or manipulated or doctored to an interest among the research community in developing
image. image forensics techniques towards identifying the trust
worthiness of digital images. Over the past decade, the
In general image forgery is the manipulation of digital image forensics emerged to help in restoring the lost trust to
images either in terms of destroying or inserting some digital images.
information in the images. An example of such forged The rest of the paper organized as follows: In section II the
image is shown in fig1. An American diplomat John Forbes classification of Image forgery is discussed, in section III
Kerry with Jane Fonda, Hollywood actress speaking to a various forgery detection techniques are discussed followed
crowd at an anti-Vietnam peace rally [1]. This is a by a summary of splicing image forgery techniques are
manipulated image by a hoaxer in trying to raise question given in section IV and end with conclusion in section V.
about John Kerry’s patriotism.
II. Classification of Image
Forgery

In literature many researches classified Image Forgery

following ways [4]. Copy-Move or region duplication
forgery is the most common image tampering technique
used because of its simplicity and effectiveness. In this type,
part of the original image is copied or moved to a destined

471 https://sites.google.com/site/ijcsis/
ISSN 1947-5500
 International Journal of Computer Science and Information Security (IJCSIS),
Vol. 14, No. 11, November 2016

location for pasting in order to hide certain details as well III. Image Forgery Detection
as duplicate parts of an image. Textured regions are used as
ideal parts for copy-move forgery, since textured area has Image forgery detection aims at verifying the authenticity
similar colour and noise variations to that of an image which of a digital image [4]. The authentication can be classified
are unperceivable for human eye looking for inconsistencies into i) Active and ii) Blind or passive approaches as shown
in image statistical properties. in fig5. Active approach includes techniques like digital
signatures or watermarking wherein a known authentication
code was embedded into the image either at the time of
creation or just before it can send through an unreliable
public channel. The authenticity can be verified by the
presence of the code with the original inserted code.
However, this method requires special hardware or software
to insert the authenticated code in the image before the
image is being used. Whereas, Blind or passive forgery
Fig2[18]: Copy-move Image forgery techniques uses the received image only for assessing its
Image splicing involves replacing of image fragments from authenticity or integrity without using any external
one or more different images on to another image in order signature or watermark of the original image. The forgery
to produce a fake image. This is one of the simple and images not leave any visual clues to indicate tampering but
commonly used tampering techniques. When splicing is leave changes its underlying statistics
performed carefully, the borders between the spliced
regions can visually be imperceptible. However, splicing
disturbs the high order Fourier statistics such as the bi-
spectrum. Image
Forensics

Active Blind

Water Digital Copy-

marking Signature Splicing
move

Pixel Camera
Based Based

Fig3 [11]: Splicing Image forgery

Image Re-sampling involves in creating a high quality Fig5: Classification of Image Forensics
forged image by applying some transformations like
rotation, scaling, stretching, skewing, flipping etc., in order The image forensic techniques classified as pixel-based
to produce a convincing composite between two objects of techniques – which detects any statistical anomalies found
different dimensions. This process requires resample the in the image at pixel level, format-based techniques - which
original image onto a new by introducing specific periodic detects any statistical correlations by a specific lossy
correlations between neighbouring pixels. compression technique, camera-based techniques - which
identifies the different artefacts produced using the camera
lens, sensor, or on-chip post processing operations, physical
based techniques - which are due to anomalies between
objects, camera and light and geometric-based techniques –
based on measurements of objects with its relative position
of camera [3].

Fig4[2]: Re-sampling image forgery

472 https://sites.google.com/site/ijcsis/
ISSN 1947-5500
 International Journal of Computer Science and Information Security (IJCSIS),
Vol. 14, No. 11, November 2016

IV. Splicing Image forgery diagonal wavelet coefficients at the highest resolution with
non-overlapping blocks. The image segmented on the basis
detection techniques of homogeneity condition into several homogenous sub-
Image splicing is a technique in which crops and pastes regions using simple region merging algorithm in order to
regions of image from the same or different image. This is detect spliced forgery. The methods work well on the image
a fundamental step used in digital photomontage, which is where there is homogeneous noise level but fails when the
very popular in digital image content editing. It is also authenticated image contains the same. As an improvement
referred as paste-up produced by sticking the images in [16] authors proposed an effective method based. First
together using available digital software tools such as the image is divided into non overlapping blocks and
Photoshop. The spliced image used in many ways such as clustering applied to make them clean and tampered blocks.
news reports, photography contest, key proof in the The detected suspicious regions are further segmented to
academic papers, and so on, which could bring certain refine noise estimation and finally applied classification to
negative influences. As the digital images become more obtain the final result. To improve the results further the
vulnerable to malicious tampering compared to their non- author’s estimate local noise variances by segmenting the
digital counterparts naturally it becomes an important and image into regions with significantly different noise
challenging research area in order to determine the variances. Simple k-means clustering algorithm applied and
authenticity of an image and detecting tampered parts of an then post-processing steps on detected regions to refine the
image. The following are various techniques found in the result. In [17] an automated technique is proposed to detect
literatures and we classify them as illumination color spliced forgery in raw images. They used relative
estimation, inconsistency in image noise levels, statistical consistency of noise parameters by looking at image
properties inherent in the source image (camera inconsistencies from quad-tree decomposition to detect the
characteristics) and other feature based methods. potential sliced images. An efficient technique is proposed
in [18] to detect region splicing. The technique is based on
Illumination color Estimation - in identifying the observed projection kurtosis concentration phenomenon.
authenticity of a digital image illumination inconsistencies The noise statistics estimation is an optimization problem
are potentially effective for splicing detection among other with closed-form solution. All these techniques based on
telltale signs. This is due to proper adjustment of the noise discrepancies in a single scale. Taking the advantage
illumination conditions is hard to achieve while creating a of multi scales as indicator for detecting spliced image
forged image. forgery [19] proposed a technique where the image
segmented into super pixels of multiple scales and then
In [12] developed a physics based on illuminant color model noise level function applied on each individual scale. Those
for detecting the difference in the local image regions. The segments which are not constrained by noise level function
authors used illumination map based on distance measure to are further processed by Optimal Parameter Combination
estimate the results there by employing in forensic analysis. Searching algorithm in order to mark the spliced regions.
This technique requires user intervention. [13] Used
inconsistencies of the illuminant color in the object region Author Year method data set works well on
in order to detect the region splicing forgeries based on local local noise
illumination estimation. They proposed to combine five Babak standard homogeneous
low-level statistics-based algorithms to estimate illuminant et al. 2009 deviation Columbia noise levels
of each horizontal and each vertical band. For further Inconsistency higher noise
development, [14] presented a new technique to detect X Pan in image Columbia, variances &
forged images of people using the illuminant color. They et al. 2012 noise levels UCID larger regions
estimated illuminant color using a statistical grey edge simple and
method and a physics-based method which exploits the specific
inverse intensity-chromaticity color space. HOG edge projection statistical
algorithm is used to combine texture and edge based cues S Lyu kurtosis Columbia, aberration by
and used machine learning late fusion. Thereby reduce user et al. 2014 concentration UCID additive noise
intervention to minimal. spliced area with
different noise
Forgeries based on inconsistency in image noise levels – variance,
noise that exists in images can be used to improve accuracy different size and
in detecting spliced image regions. It is evident that each Chi-M own different no of
image obtained by a digital camera prone to contain certain P et al. 2016 OPCS dataset spliced objects
type of noise which may happen during to process of
photons comes into the sensor until the camera output the Table1: Comparison of noise based techniques
image.

In [15] a blind forgery detection method based on local

noise inconsistencies to detect small regions corrupted by
local noise is proposed. The method uses the high pass

473 https://sites.google.com/site/ijcsis/
ISSN 1947-5500
 International Journal of Computer Science and Information Security (IJCSIS),
Vol. 14, No. 11, November 2016

Statistical properties inherent in the source image - the Other Feature based techniques - in general, any feature
consistency of inherent physics-based attributes among based techniques follows a 4 step process. Image pre-
different parts of a single image such as natural scene processing mainly to enhance the structural changes
related, imaging device properties such as camera occurred due to forgery. Feature extraction where compute
characteristics can be used in detecting forged regions of an the specific representation of data that can highlight relevant
image. information. To reduce complexity, eliminate some
insignificant features before classification. Classifier
In [20] proposed a method based on identifying the selection and modelling to identify an appropriate classifier
consistency of camera characteristics among various areas and then train set of images and fine tune the parameters.
of an image. On a segmented image, from each area a Classification discriminates the given image and classifies
camera response function (CRF) is estimated using them into two either authentic or forged. Commonly used
geometric invariants from LPIP’s. CRF cross fitting scores classifiers as SVM [5-8], KNN[23], Naïve Bayes[24],
and area intensity features are computed and given to SVM- ANN[10].
based classifier. A machine learning algorithm based on
human visual system (HVS) model is proposed in [21]. In [5] proposed a technique based on features extracted from
High correlation between spliced borders and the first few chromatic channel. After chrominance component is
fixation points obtained by edge sharpness used as visual extracted, the image divided into overlapping blocks and
cues. The visual fixation prediction algorithm is proposed LBP calculated for each block and transformed each block
to detect spliced images with visual cues. The limitation is into 2D DCT. Standard deviations corresponding DCT
that the edge sharpness cues used in this method will fail coefficients of each block are used as feature vector. A
when concealing measures, such as blur, is applied. To different perspective method is proposed in [7] where first
improve further [22] proposed a method based on an multi block discrete cosine transform (MBDCT) applied to
intrinsic camera parameter lens radial distortion for input images and apply the multi resolution LBP operator
detecting spliced image forgery where the degree of lens on the magnitude components of 2D array DCT
radial distortion across the image is used as evidence for components. Kernel Principal Component Analysis (Kernal
splicing. The algorithm measures lens radial distortion of PCA) is used to reduce the dimensionality of the feature
the image using line-based calibration. The method detects vector. In [6] authors applied multi-scale entropy filter on
works well for splicing of images when straight edges are chrominance components Cb and Cr of the input images
there but the low quality image will generate perturbations followed by LPQ operator. Feature vector is obtained by
along with straight lines which results wrong estimation of calculating the histogram of LPQ of the image with size
radial distortions. In [9] Used inconsistency in the blurriness 256. Owing to their effectiveness and simple-ness Markov
and direction of motion blur. This method cannot features [8] extracted from both DCT and DWT domains.
discriminate motion and out-of-focus blur. It can only used AEM_EDW is used to make the computational complexity
for linear motion blur and cannot be applied to the more more manageable.
complicated motion blur kernels.
Author year method data set performance
Author year method data set performance Amani LBP & CASIA
Precision- et al. 2013 DCT v1.0 97%
Yu-Feng CRF& own 70% Recall Yujin Z Kernal
Hsu et al. 2007 LPIP dataset 70% et al. 2013 PCA Columbia 90.46%
Zhenhua Qu 96.33% Saurabh Entropy CASIA
et al. 2009 HVS Columbia accuracy A et al. 2015 & LPQ v2.0 98.33%
consistency Ce.Li
of lens et al. 2015 QDCT DVMM 93.42%
H.R. radial 86%
Chennamma 2010 distortion Columbia accuracy Table3: Comparison of feature based techniques
spectral
analysis
Pravin K et of image own 93.43% V. Conclusions
al. 2011 gradients dataset accuracy
In this review we presented summary of splicing image
Table2: Comparison of Statistical properties of source
forgery techniques. The spliced images are produced from
image
different images there by the discrepancies of the image
features or camera characteristics are the main source in
detecting the forged regions of the images. Among the
pixel-based and statistical based techniques we classify
further in to illumination color estimation, statistical
characteristics of the image, noise inconsistency and finally
presented other feature based methods. There may be

474 https://sites.google.com/site/ijcsis/
ISSN 1947-5500
 International Journal of Computer Science and Information Security (IJCSIS),
Vol. 14, No. 11, November 2016

several techniques found in the literatures but, each one has “Blurred Image Region Detection And Classification,” In
its limitations. Image forensics is a burgeoning research Proc. 19th Acm Int. Conf. Multimedia, 2011, Pp. 1397–
field and despite the limitations it promises a significant 1400.
improvement in forgery detection with competition among
[12] C. Riess And E. Angelopoulou, “Scene Illumination
forgery creators and detectors. As An Indicator Of Image Manipulation,” Inf. Hiding, Vol.
6387, Pp. 66–80, 2010.
[13] Yu Fan, Philippe Carré, Christine Fernandez-
REFERENCES Maloigne” Image Splicing Detection With Local
Illumination Estimation”, ICIP 2015
[1] K L Fonda, Kerry, And P Fakery, "The Washington [14] C. Riess And E. Angelopoulou, “Exposing Digital
Post," P- A21, Feb. 2004. Image Forgeries By Illumination Color Classification”,
[2] M A Qureshi, M Deriche, “A Review On Copy Move IEEE Transactions On Information Forensics And Security,
Image Forgery Detection Techniques”, 11th IEEE Vol. 8, No. 7, July 2013
International Multi-Conference On Systems, Signals & [15] Xunyu Pan, Xing Zhang, Siwei Lyu, “Exposing Image
Devices (Ssd), 2014 Forgery With Blind Noise Estimation”, Thirteenth Acm
[3] H Farid, "Image Forgery Detection A Survey," IEEE Multimedia Workshop On Multimedia And Security, 2012
Signal Processing Magazine, Vol. 26, No. 2, Pp. 16-25, Mar [16] Xunyu Pan, Xing Zhang, Siwei Lyu, “Exposing Image
2009. Forgery With Blind Noise Estimation”, IEEE International
[4] G K Birajdar, V H Mankar, “Digital Image Forgery Conference On Computational Photography (ICCP), 2012
Detection Using Passive Techniques: A Survey”, Digital [17] Thibaut Julliand, Vincent Nozick And Hugues Talbot,”
Investigation (10) 2013 @Elsevier. Automated Image Splicing Detection From Noise
[5] Aa Alahmadi, M Hussain ; Hatim A ; Ghulam M ; Estimation In Raw Images”, 6th International Conference
George B, “Splicing Image Forgery Detection Based On On Imaging For Crime Prevention And Detection (ICDP-
Dct And Local Binary Pattern”, Global Conference On 15)
Signal And Information Processing (Globalsip), 2013 IEEE [18] Siwei Lyu ,Xunyu Pan And Xing Zhang, “Exposing
[6] Saurabh A, Satish Ch,” Image Forgery Detection Using Region Splicing Forgeries With Blind Local Noise
Multi Scale Entropy Filter And Local Phase Quantization”, Estimation”, Springer, Int J Comput Vis,2014
Internation Journal Of Image, Graphics And Signal [19] Chi-Man Pun, Bo Liu, Xiao-Chen Yuan, ”Multi-scale
Processing, 2015. noise estimation for image splicing forgery detection”,
[7] Y Zhang, Ch Zhao, Yiming Pi, Shenghong Li1, Shilin Journal Of Visual Communication Image Representation,
W”, Image-Splicing Forgery Detection Based On Local Elsevier, 2016.
Binary Patterns Of Dct Coefficients”, Security And [20]Yu-Feng Hsu And Shih-Fu Chang,” Image Splicing
Communication Networks, Published Online In Wiley Detection Using Camera Response Function Consistency
Online Library, 2013. And Automatic Egmentation”, 2007 IEEE International
[8] Ce Li1,2(&), Qiang Ma1, Limei Xiao1, Ming Li1, And Conference On Multimedia And Expo
Aihua Zhang1,” Image Splicing Detection Based On [21] Zhenhua Qu1, Guoping Qiu2, And Jiwu Huang1,”
Markov Features In Qdct Domain”, Lncs, Springer, 2015 Detect Digital Image Splicing With Visual Cues”, LNCS,
[9]P. Kakar, N. Sudha, And W. Ser, “Exposing Digital Springer, 2009.
Image Forgeries By Detecting Discrepancies In Motion [22] H. R. Chennamma, Lalitha Rangarajan, “ Image
Blur,” IEEE Trans. Multimedia, Vol. 13, No. 3, Pp. 443– Splicing Detection Using Inherent Lens Radial Distortion”,
452, Jun. 2011. International Journal Of Computer Science, 2010
[10] Zhang Z, Wang G, Bian Y, Yu Z,”A Novel Model For [23] Fahime Hakimi, Zanjan, Iran Mahdi Hariri,” Image-
Splicing Detection”, In IEEE 5th International Conference Splicing Forgery Detection Based On Improved Lbp And
On Bio-Inspired Computing: Theories And Applications K-Nearest Neighbors Algorithm”, International Journal Of
(Bic-Ta), 2010 Electronics Information & Planning, 2015.
[11] Tu K.Huynh, Khoa V.Huynh, Thuong Le-Tien, Sy [24] Muratov O, Dang-Nguyen D-T, Boato G. De Natale
C.Nguyen, “A Survey on Image Forgery Detection Fg,” Saliency Detection As A Support For Image
Techniques”, International Conference on Computing & Forensics”, 5th International Symposium On
Communication Technologies Research, Innovation, and Communications Control And Signal Processing(ISCCP),
Vision for Future (RIVF),2015.B. Su, S. Lu, And C. L. Tan, 2012

475 https://sites.google.com/site/ijcsis/
ISSN 1947-5500