detection of digital image forgery using wavelet...

4
Detection of Digital Image Forgery using Wavelet Decomposition and Outline Analysis Abhishek Kashyap, Rajesh Singh Parmar, B. Suresh, Megha Agarwal, Hariom Gupta Department of Electronics and Communication Engineering, Jaypee Institute of Information Technology, Noida-201304, Uttar Pradesh, India. Email: [email protected], [email protected], [email protected], [email protected] Abstract—Digital Images are everywhere, in newspapers, on the covers of the magazine, in courtrooms and all over the internet. Throughout the day, we are exposed to them and most of the time we trust what we see. Due to the presence of powerful editing software in the market, there is uncertainty on the credibility of the digital images as these might be manipulated, therefore we propose the methodology to identify the duplicated image based on edge analysis method. The proposed method is able to identify forged area of an image and it’s accuracy is 81.50% with in small processing time. Index Terms—Image forgery detection, Copy-move, splicing, wavelet decomposition, DCT. I. I NTRODUCTION The reliability of the images play a vital role in sensitive area such as criminal activity examination, forensic analysis, surveillance inspection and Bio-medical imaging. Digital im- ages are operated or modified very easily with the help of advanced photo editing software now a days. This makes it harder to check its genuineness of the given digital images. Hence digital contents can not be considered as a strong proof due to several types of digital image forgeries. So that it is essential to develop some analytical method to check genuineness of the digital contents. Copy-move and splicing are the most common image manipulation techniques. A region is copied from an image and pasting it to the same image at different locations for hiding important information, this process is known as copy-move forgery. In Splicing, two or more images combine and make a single composite image. To detect digital image forgery, many algorithms has been proposed [1]- [7]. In which generally two types of methods are described, such as active method and passive method. Active methods require pre-embedded information, which is inserted at the time of creation of the image but passive methods require no pre-embedded information, they work purely binary information of the digital image. Image can be characterised by some feature, which are used for pair of searching and matching process by some algo- rithms. So that we have proposed an algorithm to investigate both copy-move and splicing forgeries, when digital image is compressed. The rest of the paper is organized as follows. A review of image forgery detection is presented in section I. In Section Fig. 1. Comprehensive algorithm for proposed method II, we present proposed novel method wavelet decomposition and outline analysis for forgery detection. In section III, we provide experiments and simulation results. In section IV we present conclusions and scope of the future work. II. PROPOSED METHOD The proposed technique analyse the edge of the object in the given digital image for forgery detection. Outlines may be allowed for post-processing by blurring the sharp edges to adjust the background. Our algorithm has four important steps for forgery detection such as (i) wavelet decomposition (ii) Edge processing (iii) Outline analysis using DCT (iv) Refinement of the results, which is shown in Fig. 1. A. Wavelet decomposition This method starts with the calculation of wavelet transform of digital image. This transform provides high-high bands, low-high bands and high-low bands at different scale of the digital image. 2-band perfect reconstruction (PR) examination filter bank is achieved by one stage wavelet decomposition of a one dimensional signal, which is shown in the Fig. 2. Low pass and high pass filter are represented by h 0 and h 1 respectively [16]. By repeating this process we can get the multiple level of wavelet decomposition. Two dimensional 187 978-1-5090-2684-5/16/$31.00 ©2016 IEEE

Upload: others

Post on 31-May-2020

2 views

Category:

Documents


0 download

TRANSCRIPT

Page 1:   Detection of Digital Image Forgery using Wavelet ...static.tongtianta.site/paper_pdf/c25c658c-cc64-11e9-94da-00163e08bb86.pdfboth copy-move and splicing forgeries, when digital

Detection of Digital Image Forgery using WaveletDecomposition and Outline Analysis

Abhishek Kashyap, Rajesh Singh Parmar, B. Suresh, Megha Agarwal, Hariom GuptaDepartment of Electronics and Communication Engineering,

Jaypee Institute of Information Technology,Noida-201304, Uttar Pradesh, India.

Email: [email protected], [email protected],[email protected], [email protected]

Abstract—Digital Images are everywhere, in newspapers, onthe covers of the magazine, in courtrooms and all over theinternet. Throughout the day, we are exposed to them and mostof the time we trust what we see. Due to the presence of powerfulediting software in the market, there is uncertainty on thecredibility of the digital images as these might be manipulated,therefore we propose the methodology to identify the duplicatedimage based on edge analysis method. The proposed method isable to identify forged area of an image and it’s accuracy is81.50% with in small processing time.

Index Terms—Image forgery detection, Copy-move, splicing,wavelet decomposition, DCT.

I. INTRODUCTION

The reliability of the images play a vital role in sensitivearea such as criminal activity examination, forensic analysis,surveillance inspection and Bio-medical imaging. Digital im-ages are operated or modified very easily with the help ofadvanced photo editing software now a days. This makes itharder to check its genuineness of the given digital images.Hence digital contents can not be considered as a strongproof due to several types of digital image forgeries. Sothat it is essential to develop some analytical method tocheck genuineness of the digital contents. Copy-move andsplicing are the most common image manipulation techniques.A region is copied from an image and pasting it to the sameimage at different locations for hiding important information,this process is known as copy-move forgery. In Splicing, twoor more images combine and make a single composite image.To detect digital image forgery, many algorithms has beenproposed [1]- [7]. In which generally two types of methods aredescribed, such as active method and passive method. Activemethods require pre-embedded information, which is insertedat the time of creation of the image but passive methodsrequire no pre-embedded information, they work purely binaryinformation of the digital image.

Image can be characterised by some feature, which are usedfor pair of searching and matching process by some algo-rithms. So that we have proposed an algorithm to investigateboth copy-move and splicing forgeries, when digital image iscompressed.

The rest of the paper is organized as follows. A review ofimage forgery detection is presented in section I. In Section

Fig. 1. Comprehensive algorithm for proposed method

II, we present proposed novel method wavelet decompositionand outline analysis for forgery detection. In section III, weprovide experiments and simulation results. In section IV wepresent conclusions and scope of the future work.

II. PROPOSED METHOD

The proposed technique analyse the edge of the object inthe given digital image for forgery detection. Outlines maybe allowed for post-processing by blurring the sharp edgesto adjust the background. Our algorithm has four importantsteps for forgery detection such as (i) wavelet decomposition(ii) Edge processing (iii) Outline analysis using DCT (iv)Refinement of the results, which is shown in Fig. 1.

A. Wavelet decomposition

This method starts with the calculation of wavelet transformof digital image. This transform provides high-high bands,low-high bands and high-low bands at different scale of thedigital image. 2-band perfect reconstruction (PR) examinationfilter bank is achieved by one stage wavelet decomposition ofa one dimensional signal, which is shown in the Fig. 2.

Low pass and high pass filter are represented by h0 andh1 respectively [16]. By repeating this process we can getthe multiple level of wavelet decomposition. Two dimensional

187978-1-5090-2684-5/16/$31.00 ©2016 IEEE

Page 2:   Detection of Digital Image Forgery using Wavelet ...static.tongtianta.site/paper_pdf/c25c658c-cc64-11e9-94da-00163e08bb86.pdfboth copy-move and splicing forgeries, when digital

Fig. 2. A stage of wavelet decomposition [16]

Fig. 3. 3-stage Subbands, 2-D wavelet decompostion [16]

wavelet decomposition can be achieved bay extended versionof one dimensional wavelet decomposition [17], [18].

M stage decomposition, two dimensional wavelet pro-vides a low frequency band LLM and high frequency bands(HLj , LHj , HHj , j = 1, 2, ,M) as shown in Fig.3.

Correlation among wavelet coefficient is reduced by wavelettransform and this process allows to compact the energy ofthe input into a small number of coefficients [19]. Lowerfrequency band LLM contain the more energy in comparedto other high frequency bands. Using Harr wavelet ψ(z), weare analysing coarse part of digital image. Where ψ(z) isorthogonal to scaling function φ(z) [21] – [23]. It is givenby the eq. (1)

ψ (z)=

∞∑p=−∞

(−1)paN−1−p√2∅ (2z− p). (1)

Wavelet decomposition function in two dimensions g(z, y)is defined [23] by the eq. (2)

g (z, y) =∞∑

q=−∞

∞∑p=−∞

∞∑l=−∞

dq,p,lψq,p (z)ψq,l (y) . (2)

B. Edge processing

Edge is located for further investigation as appeared in Fig.1. In this step we have utilized canny edge detector for thispurpose. For edge block determination, R × R must containat least one pixel.

C. Outline analysis using DCT

For processing cost-coefficient reduction into three classes(i.e. high frequencies HFs, middle frequencies MFs and lowfrequencies LFs), DCT transform is applied to an edge block.high frequencies HFs describe image texture and image detail,low frequencies LFs give large scale feature and outline of theimage. The feature classification is achieved by calculatingthe ratio of high frequency power to low frequency power(HF/LF ) with the help of K-mean clustering.

D. Refinement of the results

Forged region, which is detected in this process may containsome block, which is genuine. To suppress the false alarm dis-tance of each forged block to other forged block is measuredby using minimum euclidian distance.

The similarity measure S(Bm, Bn) [15] between two sub-blocks is defined as:

S (Bm, Bn) =1

1 + ρ (Bm, Bn)(3)

Where ρ is Euclidean distance between two sub-blocks:

ρ (Bm, Bn) =

(dim∑k=1

(Bm [k]−Bn [k])2

)1/2

(4)

If S(Bm, Bn) ≥ T , (where T is the minimum requiredsimilarity), then we have analyzed the neighborhood of Bm

and Bn. Threshold (T ) played a very important role to measurethe degree of reliability between sub-blocks m and n, whichis used to take a decision for forged digital images. If thereare more blocks in one area means more probability to bemanipulated, while single detected block in a large area havingless probability to be forged.

III. EXPERIMENTAL RESULTS

Copy-move and splicing forgery both are detected by pro-posed method. Fig. 4, 5, 6 and 7 show the output of ourproposed algorithm, when this algorithm is applied to theforged images. Receiver Operating Characteristic (ROC) curvefor the proposed method is shown in Fig. 8. The proposedmethod shows the best performance compared to other one,which is shown in the ROC curve. Sensitivity and specificityfor the proposed method with respect to different quality factoris shown in Fig. 9. comparison between different methods andthe proposed method is shown in table I. R. M. Rad method[24] has the detection accuracy (78.80%) and our method hasthe detection accuracy (81.50%). Therefore proposed methodis most suitable for digital image manipulation problems.

188

Page 3:   Detection of Digital Image Forgery using Wavelet ...static.tongtianta.site/paper_pdf/c25c658c-cc64-11e9-94da-00163e08bb86.pdfboth copy-move and splicing forgeries, when digital

Fig. 4. (i) Original image (ii) Duplicated image (iii) Forged region map

Fig. 5. (i) Original image (ii) Duplicated image (iii) Forged region map

TABLE ICOMPARISON OF PRESENT WORK WITH SOME RECENT WORKS

Features TP(%) TN(%) Accuracy(%)Lyu and Farids [11] 78.20 69.39 73.75Shi et al.s [12] 75.55 76.02 75.78Zou et al.s [13] 77.40 75.07 76.21R. M. Rad (SVD) [24] 76.90 74.33 75.64R. M. Rad (SVD-DCT) [24] 77.56 77.63 77.60R. M. Rad (SVD+SVD-DCT) [24]

80.11 77.61 78.80

Proposed method 83.33 76.0 81.50

IV. CONCLUSIONS

Proposed method is based on single framework to examinesplicing and copy-move type of forgery in the digital images.

Fig. 6. (i) Original image (ii) Duplicated image (iii) Forged region map

Fig. 7. (i) Original image (ii) Duplicated image (iii) Forged region map

Fig. 8. Performance analysis curve (ROC curve)

Fig. 9. Sensitivity and specificity vs Quality factor

Object’s outline is analysed in terms of smoothness andsharpness in this method. It is clear from experimental resultsthat proposed technique can successfully detect and localizeforgeries in digital images. Hence edge analysis method isan appropriate method for image forgeries such as splicingand copy-move. In future some other techniques can also bedevelop, when small discrepancies are not detectable afterforged region has been scaled or rotated.

REFERENCES

[1] T. Qazi, K. Hayat, S. U. Khan, S. A. Madani, I. A. Khan, J. Kolodziej,H. Li, W. Lin, K. C. Yow, C. Z. Xu, “Survey on Blind Image ForgeryDetection,” in IET Image Processing, vol. 7, no. 7, pp. 660-670, 2013.

[2] Zh. Lin,J. He,X. Tang and Ch. K. Tang, “Fast, Automat. and FinegrainedTampered JPEG Image Detection via DCT Coefficient Anal.,” PatternRecognition, vol. 42, no. 11, pp. 2492-2501, 2009.

[3] W. Wang, J. Dong, and T. Tan, “Exploring DCT Coefficient QuantizationEffects for Local Tampering Detection,” IEEE Trans. on Inform. Forensicsand Security, vol. 9, no. 10, pp. 1653-1666, 2014.

[4] G. Lynch, F. Y. Shih and H.-Y. M. Liao, “An Efficient Expanding BlockAlgorithm for Image Copy-move Forgery Detection,” Inform. Sci., vol.239, no. 0, pp. 253-265, 2013.

[5] T. Bianchi, A. De Rosa and A. Piva, “Improved DCT Coefficient Analysisfor Forgery Localization in JPEG Images,” in Proc. IEEE Int. Conf. Acoust,Speech and Signal Process. (ICASSP), pp. 2444- 2447, 2011.

189

Page 4:   Detection of Digital Image Forgery using Wavelet ...static.tongtianta.site/paper_pdf/c25c658c-cc64-11e9-94da-00163e08bb86.pdfboth copy-move and splicing forgeries, when digital

[6] V. Christlein, C. Riess, J. Jordan, C. Riess, and E. Angelopoulou, “AnEvaluation of Popular Copy-Move Forgery Detection Approaches,” IEEETrans. on Inf. Forensics and Security, vol. 7, no. 6, pp. 1841-1854, 2012.

[7] J. Li, X. Li, B. Yang, and X. Sun, “Segmentation-Based Image Copy-Move Forgery Detection Scheme,” IEEE Trans. on Inf. Forensics andSecurity, vol. 10, no. 3, pp. 807-518, 2015.

[8] G. Schaefer, and M. Stich, “UCID - An Uncompressed Colour ImageDatabase,” in Proc. SPIE of Storage and Retrieval Methods and Applicat.for Multimedia, pp. 472-480, 2004

[9] Tian-Tsong Ng; Shih-Fu Chang, “A model for image splicing,” in Proc.IEEE Int. Conf. Image Processing (ICIP), Singapore, Oct. 24-27, 2004,Vol.2, pp. 1169-1172.

[10] Fu, Dongdong, Yun Q. Shi, and Wei Su, “Detection of image splicingbased on Hilbert-Huang transform and moments of characteristic functionswith wavelet decomposition,” in Proc. Digital Watermarking, SpringerBerlin Heidelberg, 2006, pp. 177-187.

[11] Lyu, S. and H. Farid, “Detecting hidden messages using higherorderstatistics and support vector machines,” Information Hiding, Springer,2003.

[12] Shi, Y.Q., et al., “Image steganalysis based on moments of characteristicfunctions using wavelet decomposition, prediction-error image, and neuralnetwork,” ICME, Citeseer, 2005.

[13] Zou, D., et al., “Steganalysis based on Markov model of thresholdedprediction-error image,” in IEEE International Conference on Multimediaand Expo, 2006.

[14] Z. Moghaddasi, H. A. Jalab and R. M. Noor, “SVD-based image splicingdetection,” in International Conference on Information Technology andMultimedia (ICIMU), Putrajaya, 2014, pp. 27-30.

[15] Kashyap, Abhishek; Joshi, Shiv Dutt, “Detection of copy-move forgeryusing wavelet decomposition,” in Proc. IEEE Int. Cof. Signal Processingand Communication (ICSC), Noida, Dec.12-14, 2013, pp.396-400.

[16] Jiwani, L.K.; Joshi, S.D.; Visweswaran, G.S., “Spectral Density DrivenWavelet Representation of 2-D Images,” in Proc. IEEE Int. SymposiumSignal Processing and Information Technology, Aug. 2006, pp. 138-143.

[17] Mallat, S.G. “A theory for multiresolution signal decomposition: thewavelet representation,” IEEE Trans. Pattern Anal. Machine Intell., vol.11,no.7, pp.674,693, Jul 1989.

[18] A. Jain, Fundamentals of Digital Image Processing. Englewood Cliffs,NJ: Prentice Hall, 1989.

[19] Usevitch, B.E.; Orchard, M.T., “Smooth wavelets, transform coding,and Markov-1 processes,” IEEE Trans. Signal Processing, vol.43, no.11,pp.2561,2569, Nov 1995.

[20] Kashyap, Abhishek; B. Suresh; Agrawal, Megha; Gupta, Hariom; Joshi,Shiv Dutt, “Detection of splicing forgery using wavelet decomposition,” inProc. IEEE Int. Conf. Computing, Communication and Automation (ICCCA2015), 15-16 May 2015.

[21] I. Daubechies, “Ten Lectures on Wavelets,” in Proc. Reg. Conf. CBMS-NSF SIAM Series in Applied Mathematics, Philadelphia, PA, 1992, vol.61.

[22] I. Daubechies, “Orthonormal bases of compactly supported wavelets,”Commun. Pure Appl. Math., vol. 41, no. 7, pp. 909996, October 1988.

[23] John R. Williams and Kevin Amaratungay, “Introduction to Wavelets inEngineering,” International Journal for Numerical Methods in Engineer-ing, vol. 37, no. 14, pp. 23652388, July 1994.

[24] R. M. Rad and K. Wong, “Digital image forgery detection by edgeanalysis,” in IEEE International Conference on Consumer Electronics -Taiwan (ICCE-TW), Taipei, 2015, pp. 19-20.

190