A Copy-move Image Forgery Detection Based on Speeded Up Robust Feature Transform and Wavelet

Transforms

Mohammad Farukh Hashmi1, Vijay Anand2, Avinash G. Keskar3 1,2,3Department of Electronics and Communication Engineering

Visvesvaraya National Institute of Technology Nagpur, 440010 (India)

farooq78699@gmail.com, vijjanand117@gmail.com, agkeskar@ece.vni.ac.in

Abstract— We are currently leaving in a pure digital world where all type of information is mostly stored in digital form. Thus storage of the information is no more a problem and the information can be easily passed on from one place to another in digital format. The big benefit however comes with the hidden loss and in this case it is the tampering of images and videos which has become a matter of serious concern in recent days because of the readily available software tools in the market like Photoshop etc. which can be used by a common person to tamper the image or video for hiding or changing the original contents. Thus for the aforementioned problem, we in this paper propose a series of algorithms which are combination of speeded-up robust feature transforms and Wavelet Transforms. In doing so we will first discuss Speeded-Up Robust Feature (SURF), SURF in combination with Discrete Wavelet Transform (DWT), SURF in combination with Dyadic Wavelet Transform (DyWT). These algorithms are different from the previously proposed algorithm in the manner that they are applied on the entire image to extract features rather than dividing the image into the blocks. From the results obtained we are able to conclude the proposed algorithms are better than their counterparts both in terms of computational complexity and invariance to scale and rotation and also for the combination of attacks.

Keywords— Digital Image forgery, DyWT (dyadic wavelet transform), DWT (Discrete Wavelet Transform), SUFR (Speeded-Up Robust Feature).

I. INTRODUCTION In this complete digital world “seeing is no more

believing”. Most of the information is carried in a digital form especially in the form of either digital images or digital videos. Thus, they form the main stream of the information carrier. These sources can be manipulated very easily. In this paper, we will focus on image forgery, which has become a topic of serious concern. The image editing software such as Adobe Photoshop is readily available using which any given image can be easily doctored, which can lead to serious consequences, as these tampered images can be presented as a part of evidence in the court room leading to a wrong decision and creating the false belief in many real-world applications.

Therefore the issue of authentication of the images has to be taken very seriously.

Most of the forgery detection techniques are categorized into two major domains: intrusive/non-blind and non-intrusive/blind [1] as shown in the Fig.1.

Figure 1. Classification of image forgery detection technique.

Intrusive method which is also known as a non-blind method requires some digital information to be embedded in the original image when it is generated, and thus it has a limited scope. Some of the examples of these methods are watermarking and using digital signature of the camera. And since not all the digital devices can provide these features which leads to the failure of intrusive methods. On the other hand, non-intrusive method which is also known as a blind method does not require any embedded information. A digital image is said to be forged when its original version is tampered

Digital Image Forgery Detection

Intrusive/Non Blind Methods

Non-Intrusive/Blind Methods

Digital signature Digital watermarking

Forgery Type Dependent

Forgery Type Independent

Copy-move Detection

Image splicing

Detection

Re-sampling Detection

Compression Detection

2014 5th International Conference on Computer and Communication Technology

978-1-4799-6758-2/14/$31.00 ©2014 IEEE 147

by applying various transformations like that of rotation, scaling, resizing, etc. It may also happen that an image is tampered by adding noise or by removing or adding some objects to hide the real information [1]. Most commonly used image forgery method is copy-move forgery in this a part of original image is copied and pasted on other parts once or may be multiple times to hide some information. The ease and effectiveness of copy-move forgery makes it the most common forgery that is used to alter the content of an image [2]. A copy-move forgery example is presented in Fig.2 in which (a) part denotes the original image and (b) denotes the forged image with only one truck in it.

(a) Original image (b) Forged Image

Figure 2. An example of Copy-Move Forgery.

Currently, we are focusing on copy-move forgery detection because the task of tamper detection becomes more difficult in this case. This is because when we copy a region from an image and paste it on that very image they have similar characteristics of that of the original image. We have used dyadic wavelet transform (DyWT) to decompose an image. DyWT is shift invariant and is better than discrete wavelet transform (DWT) for data analysis [3]. After the decomposition we apply SURF algorithm to extract the features, it obtains descriptor for colour as well as boundary images as wavelet produces both. We now perform a searching to search for occurrence of same features at different part of images. Image blocks that return similar SURF features from all four images are marked as forged regions.

The remaining paper is presented in following manner. Next section deals with all previous work related to image forgery detection. Section III completely explains the proposed method, section IV deals with simulation results and evaluation of performance parameter and in the end, we have conclusion and references.

II. RELATED WORK The main aim of copy-move forgery is to detect the different copied regions and their pasted one. This is not an easy job because the copied part has mostly the same characteristic as that of original image and there are chances that these copied parts are processed by operation like noise addition, filtering and geometrical distortion. Thus in order to correctly detect the copied region forgery detection techniques should detect the copied region even if they are slightly different. In this section, we have reviewed most of the blind methods of copy-move image forgery detection. Most of the previously proposed algorithm for image forgery detection were block

based in which the given image is first divided into overlapping blocks and then similarity between the blocks are obtained to conclude the image forgery. And then we go for reviewing the algorithms based on feature extraction. As we are dealing with the invariant feature transforms in this paper, these algorithms are different from the previously proposed one because these are not block based instead; they are applied on the complete image for extracting the features. Mainly this concept is achieved by methods like SIFT and SURF. These are used to extract distinctive features from the image and from those feature they generate the feature descriptor which are invariant to scaling, rotation etc. Muhammad et al. [3]-[4] presented a blind and robust technique using dyadic (un-decimated) wavelet transform (DyWT) which has better results than Discrete Wavelet Transform (DWT).Huang et al. [5] used SIFT algorithm for the first time to represent the features of the given image. In this algorithm matching between the SIFT key points are obtained by using best-bin-first nearest-neighbor approach. SIFT algorithm is invariant to changes in illumination, rotation, scaling, etc. SIFT was adopted by Ardizzone et al. [6] for detecting multiple copies in copy move image forgery. Zhang et al. [7] first extracted the key points and then found the matching between them to obtain key points pairs. Author in order to differentiate between the source and target used a strategy based on match vector orientations. Amerini et al. [8], deals with detecting whether an image has been forged or not specially using copy-move forgery by using Scale Invariant Features Transform (SIFT). SIFT allow us to understand that copy-move forgery has occurred, and it also recovers from geometric transformation used for cloning. Using this method we can also deal with multiple copy-move forgery. Hashmi et al. [9] provided an algorithm by combining DWT and SIFT to detect copy-move image forgery in which proposed algorithm has better results than the previously existing one. Anand et al. [10] proposed an algorithm to detect the digital image copy move forgery to overcome the sustained attacks using SIFT and DyWT methods. In this DyWT which is shift invariant is combined with SIFT to obtain the forged part and results obtained were better than SIFT alone and algorithm proposed in [9]. Xu Bo et al. [11] presented a fast method for detecting image forgery using SURF descriptors which are scale and rotation invariant. Euclidean distance is used to find the nearest neighbor and thus to obtain matching. Region duplication forgery in images is detected by shivakumar et al. [12] using SURF. KD-Tree is used for multidimensional data matching. Input image on which SURF is applied to obtain features and then KD-Tree is used for finding matching and obtaining the duplicated region in the image. Mishra et al. [13] has proposed region duplication forgery detection using SURF and hierarchical agglomerative clustering (HAC). On the test image SURF is applied to obtain the key points and there corresponding descriptors and HAC is performed on the matched key points to obtain copy-move forgery detection. Al-Qershi et al. [14] provided a state-of-art of passive detection of copy-move forgery in digital images. The key current’s issues are discussed for developing robust passive

148

copy-move forgery detection. Leida Li et al. [15] proposed an efficient method using local binary patterns, in this image is first image is filtered and then divided into over-lapping circular blocks; features of these blocks are calculated using local binary patterns to detect the forged regions. Birajdar et al [16] summarized a complete survey on digital image forgery detection using passive techniques, which discussed currently available forgery detection techniques and also provide further recommendation for future research. Mahalakshmi et al [17] provided detection of digital image forgery by exploring basic image manipulations done on the images. Here they have presented techniques to detect image is manipulated using basic method like copy-move, region duplication, splicing etc. In [23], Zhang has presented a method of detecting forgery in flat regions using SURF, the proposed method can also be to detect flat region in images in an effective way, and extract FMT features after blocking the region. By using matching algorithms of similar blocked images, image forgeries in flat region can be determined, which results in the completing of the entire image tamper detection.

III. PROPOSED METHODOLOGY Through this paper, we are going to propose a new technique for copy-move forgery detection. First image is transformed into wavelet domain and SURF is applied on the transformed image to obtain the features. As wavelet produces multispectral components, features are more predominant [18]. Thus after obtaining interest point feature descriptor we go for finding matching between these feature descriptors to conclude whether tampering is done to the given image or not. Our works confirm that SURF features are an optimal solution because of their high computational efficiency and robust performance.

A. SURF (Speeded-Up Robust Feature) SURF was first presented by Bay et al. [19] and is used to obtain scale and rotation invariant detector as well as descriptor. SURF is better than previously proposed methods in terms of repeatability, distinctiveness, and above all robustness. Main reason why SURF became more popular is that its computation time is much faster than any other schemes. SURF generally consists of detection of interest point and interest point descriptor generation.

a) Interest Point Detection b) Scale Space Representation c) Orientation Assignment d) Descriptor Generation e) Feature Matching

B. SURF combined with DWT In this section we are going to propose a new algorithm which combines SURF with DWT. Here first DWT is applied on the

given image to decompose the image into four sub-images. LL, LH, HL, HH and out of these four sub-images LL part contains most of the information. So we apply SURF on the LL part to obtain key points and there corresponding descriptors. And in the end we go for finding matching between descriptor using best bin first method and also key point clustering is performed using single linkage, average linkage, and ward linkage to obtain the duplicated region. The complete flow of the algorithm is presented in the Fig.3.

Figure 3. Block diagram of proposed algorithm combining SURF and DWT.

C. SURF combined with DyWT As proposed in the previous section where we combined SURF with DWT and seen some positive results as compared to SURF alone itself, so we decided to go for DyWT in place of DWT. Thus in this section we propose a new algorithm combining SURF and DyWT. Overall process is similar to the above mentioned process except that we use DyWT in place of DWT. As we have learned from the literature reviews that DyWT has better performance when it comes to data analysis and also DyWT is shift invariant thus applying DyWT does not down sample the image at every stage thus information is intact and then we apply SURF to extract the key points and their corresponding descriptors. And in the end we go for finding match between these descriptor vectors to mark the forged region. Complete flow of the process is similar with that of SURF combined with DWT except for that of DWT we have to use DyWT and the remaining flow remains the same.

Given image

Apply DWT

Actual Horizontal

Vertical Diagonal

Apply SURF on Actual part for Key-points detection and feature extraction

Obtain SURF Feature Descriptor Vector

Find match between the descriptor vectors

Mark the forged regions

149

IV. TEST OF ROBUSTNESS AGAINST VARIOUS ATTACKS

A. DWT +SURF Through this paper we present a new method of detecting copy move image forgery using SURF with DWT. As shown in the flow chart above first DWT is applied on the test image and then we apply SURF on the LL part of the image. Since SURF will provide us with key points and there corresponding descriptor vectors among which we have to find similarity in order to conclude whether the forgery has been done or not. The complete process is represented in Fig.4.

(a) Rotation Attack (b) Rotation to higher degree

(c) Scaled to lower scale (d) Scaled version

(e) Rotation and Scale (f) Simple copy-move

Figure 4. Images showing detection of copy-move forgery under various attacks using algorithm based on DWT and SURF.

B. DyWT +SURF We in this paper presented one more new approach for detecting copy move image forgery using DyWT and SURF. As DyWT is superior to DWT in terms of data analysis and also it is shift invariant and thus when combined with SURF will generate a more robust method for image forgery detection. In this section we will present the complete flow of

the process through series of images and also robustness of the proposed algorithm. The proposed algorithm is applied on standard database of MICC-F220 images and results obtained are presented in Fig.5.

(a) Simple copy-move (b) Rotation Attack

(c) Rotation to higher degree (d) Higher Rotation and Scaling

(e) Scaling Attack (f) Rotation and Scale Figure 5. Images showing detection of copy-move forgery under various attacks using algorithm based on DyWT and SURF.

V. SIMULATION RESULTS AND EVALUTION OF PERFORMANCE PARAMETERS

In result analysis, we applied the proposed algorithm over a standard dataset MICC-F220 [8]. This simulation has been performed on MATLAB 2012a software with 32GB Ram and core i7 processor. DyWT of the test image is first calculated and on the low-frequency component of DyWT we apply SURF to extract the features which are nothing but descriptor vectors of the object of interest in the test image and the final step is to go matching these features to detect copy- move forgery. We are focusing on two major performance parameters at the image level to determine the fact that an image has been tampered or not. Considering the image level some of the important measures are described in Table I.

150

TABLE I. EVALUTION MEASURES DESCRIPTION

Evalution Measures Description

True Positive (TP) Number of images that have been correctly detected as forged

False Positive (Fp) Number of images that have been falsely detected as forged

False Negative (FN) Number of images that have been falsely missed but they are

forged.

True Negative (TN) Number of images that have been correctly detected as not-forged

(clean)

From the above mentioned measures we calculated Precision, p, and Recall, r which are defined as :

P

P P

TpT F

=+

and P

P N

TrT F

=+

(1)

Precision repersents the probability that detected forgery is truly a forgery and Recall repersents probability that a forged image has been detected. Recall is also called as True Positive Rate (TPR) [22].Thus after performing experiment on database of MICC F220, we tabulated the result obtained in Table II.

TABLE II. PERFORMANCE EVALUTION OF SURF BASED ALGORITHM

Parameters

TP

TN

FP

FN

Precision

(p)

Recall

(r)

False Positive

Rate (FPR)

SURF 77 88 22 33 78% 70% 20%

DWT+SURF 71 88 21 39 77% 64% 19%

DyWT+SURF 84 84 25 26 77% 76% 23%

So from the above tables and chart we can conclude that SURF when combined DyWT and DWT provide better precision rate and recall rate. Thus from the above results we came to the conclusion that we can use SURF combined with DWT And DyWT for detecting copy move image forgery detection.

Comparison of our proposed algorithm with the previously proposed algorithm is presented in Table III.

TABLE III. COMPARISON WITH PREVIOUS ALGORITHMS

Methods True Positive Rate (TPR) or Recall (r)

Amerini et al. [8] 100% Mishra et al. [13] 73.64% DWT + SURF 64% DyWT +SURF 76%

Thus as seen from above mentioned tables we can conclude that our proposed algorithm which combines DyWT and SURF has better precision rate and also our TPR is better than

SURF results obtained in [13]. Graphical representation of above comparison between various methods with respect to TPR and FPR is presented in Fig.6.

0

20

40

60

80

100

120

Amerini et al. [8]

Mishra et al. [13]

DWT + SURF

DyWT +SURF

TPR

FPR

Figure 6. Graphical representation of performance measure of various algorithms.

VI. CONCLUSION Through this paper we have covered almost all the algorithms related SURF. As we can make out from the related work section that SIFT is mostly used invariant feature transform than SURF, but both have their own advantages. SURF is faster than SIFT and descriptor vector dimension of SURF is smaller than that of SIFT. In this paper we have presented algorithms which combine invariant feature transforms with DWT and also with DyWT. The main objective behind this approach is to obtain a unique and more robust technique to detect copy move image forgery and also it should be able to sustain various attacks. From the results obtained we can conclude that the proposed algorithm has better precision rate as well as recall rate thus both the newly proposed algorithm in this paper are feasible one.

ACKNOWLEDGMENT We would like to thanks our Supervisor and Coordinator of Center of Excellence Dr. Avinash G. Keskar for his constant encouragement and guidance toward this project. This project is funded from Centre of Excellence (CoE), Department of Electronics Engineering, VNIT Nagpur. Special thanks to Director VNIT Nagpur for providing institutional facilities and needed administrative and authoritative support during the work at VNIT.

REFERENCES [1] Muhammad, Najah, Muhammad Hussain, Ghulam Muhammad, and

George Bebis."Copy-move forgery detection using dyadic wavelet transform." In Proceedings of IEEE Eighth International Conference on Computer Graphics, Imaging and Visualization (CGIV-2011), pp. 103-108, 2011.

[2] Jing, Li, and Chao Shao. "Image Copy-Move Forgery Detecting Based on Local Invariant Feature." Journal of Multimedia, vol. 7, no. 1, pp.90-97, 2012.

151

[3] Muhammad, Ghulam, Muhammad Hussain, and George Bebis. "Passive copy move image forgery detection using undecimated dyadic wavelet transform." Digital Investigation, vol. 9, no. 1, pp.49-57, 2012.

[4] Muhammad, Ghulam, Muhammad Hussain, Khalid Khawaji, and George Bebis. "Blind copy move image forgery detection using dyadic undecimated wavelet transform." In Proceedings of IEEE 17th International Conference on Digital Signal Processing (DSP-2011), pp. 1-6, 2011.

[5] Huang, Hailing, Weiqiang Guo, and Yu Zhang, "Detection of copy-move forgery in digital images using SIFT algorithm." In Proceedings of Pacific-Asia Workshop on Computational Intelligence and Industrial Application (PACIIA-2008), vol. 2, pp. 272-276 , 2008.

[6] E. Ardizzone, A. Bruno, and G. Mazzola, “Detecting multiple copies in tampered images.” In Proceedings of the 17th IEEE International Conference on Image Processing (ICIP -10), pp.2117–2120, September 2010.

[7] Zhang, Chenyang, Xiaojie Guo, and Xiaochun Cao. "Duplication localization and segmentation." In Proceedings of Advances in Multimedia Information Processing(PCM 2010), pp. 578-589, Springer Berlin Heidelberg, 2010.

[8] Amerini, Irene, Lamberto Ballan, Roberto Caldelli, Alberto Del Bimbo, and Giuseppe Serra. "A sift-based forensic method for copy–move attack detection and transformation recovery." IEEE Transactions on Information Forensics and Security, vol. 6, no. 3, pp. 1099-1110, 2011.

[9] Hashmi, Mohammad Farukh, Aaditya R. Hambarde, and Avinash G. Keskar. “Copy Move Forgery Detection using DWT and SIFT Features.” In Proceedings of 13th IEEE International Conference on Intelligent Systems Design and Applications (ISDA-2013), pp.188-193, December 2013.

[10] Anand, Vijay, Mohammad Farukh Hashmi, and Avinash G. Keskar. "A Copy Move Forgery Detection to Overcome Sustained Attacks Using Dyadic Wavelet Transform and SIFT Methods." In Proceedings of the 6th Asian Conference on Intelligent Information and Database Systems (ACIIDS 2014), Springer International Publishing, pp. 530-542, 2014.

[11] Bo, Xu, Wang Junwen, Liu Guangjie, and Dai Yuewei. "Image copy-move forgery detection based on SURF." In Proceedings of IEEE International Conference on Multimedia Information Networking and Security (MINES-2010), pp. 889-892, 2010.

[12] Shivakumar, B. L., and S. Santhosh Baboo. "Detection of Region Duplication Forgery in Digital Images Using SURF." International Journal of Computer Science Issues (IJCSI), vol. 8, no. 4, pp.199-205, 2011.

[13] Mishra, Parul, Nishchol Mishra, Sanjeev Sharma, and Ravindra Patel. "Region Duplication Forgery Detection Technique Based on SURF and HAC." The Scientific World Journal, vol. 2013, Article ID 267691, pages 8, 2013.

[14] Al-Qershi, Osamah M., and Bee EeKhoo. "Passive detection of copy-move forgery in digital images: State-of-the-art." Forensic Science International, vol. 231, no. 1, pp. 284-295, 2013.

[15] Li, Leida, Shushang Li, Hancheng Zhu, Shu-Chuan Chu, John F. Roddick, and Jeng-Shyang Pan. “An Efficient Scheme for Detecting Copy-move Forged Images by Local Binary Patterns”, Journal of Information Hiding and Multimedia Signal Processing, vol. 4, no. 1, pp. 46-56, January 2013.

[16] Birajdar, Gajanan K., and Vijay H. Mankar. "Digital image forgery detection using passive techniques: A survey." Digital Investigation, vol.10, no. 3, pp. 226–245, 2013

[17] Devi Mahalakshmi, S., K. Vijayalakshmi, and S. Priyadharsini. “Digital image forgery detection and estimation by exploring basic image manipulations” Digital Investigation, vol. 8, no. 3, pp. 215–225, 2012.

[18] Amerini, Irene, Lamberto Ballan, Roberto Caldelli, Alberto Del Bimbo, Luca Del Tongo, and Giuseppe Serra. "Copy-move forgery detection and localization by means of robust clustering with J-linkage." Signal Processing: Image Communication, vol. 28, pp.659–669, April 2013.

[19] Bay, Herbert, Tinne Tuytelaars, and Luc Van Gool. "Surf: Speeded up robust features." In Proceedings of 9th European Conference on Computer Vision (ECCV 2006), Springer Berlin Heidelberg ,pp. 404-417, 2006.

[20] Lowe, David G., "Distinctive image features from scale-invariant key points", International Journal of Computer Vision, vol. 60, no. 2, pp. 91-110, 2004.

[21] Bay, Herbert, Andreas Ess, Tinne Tuytelaars, and Luc Van Gool. "Speeded-up robust features (SURF)." Computer Vision and Image Understanding, vol. 110, no. 3 pp.346-359, 2008.

[22] Christlein, Vincent, Christian Riess, Johannes Jordan, and E. Angelopoulou. "An evaluation of popular copy-move forgery detection approaches.",IEEE Transactions on Information Forensics And Security, vol. 7, no. 6, pp.1841-1854, December 2012.

[23] Zhang, Guang-qun, and Hang-jun Wang. "SURF-based Detection of Copy-Move Forgery in Flat Region." Interrnational Journal of Advancements in Computing Technology(IJACT), vol. 4, no. 17, pp.521-529 , September 2012.

152