ASRD: Algorithm for Spliced Region Detection in Digital Image Forensics

Conference paper
Part of the Advances in Intelligent Systems and Computing book series (AISC, volume 575)

Abstract

Image splicing is one of the most frequently exercised in the area of image forgery that is quite challenging to be identified. After reviewing existing techniques towards identification of spliced region, it was found that existing techniques are either computationally expensive or do not address the cumulative problem. Hence, this paper, a novel and simple algorithm is presented called as ASRD i.e. Algorithm for Spliced Region Detection. A simple statistical-based approach is presented that perform partitioned blocks followed by detection of various artifacts among the neighbor blocks. The algorithm then implicates a classification condition for tampered and non-tampered region to truly identify the spliced region. With an aid of histogram analysis, true positive score, true negative score, accuracy and computational performance, the proposed algorithm was found to excel better performance in detection of spliced region.

Keywords

Image splicing Image forensics Color filter array Localization Accuracy 

1 Introduction

The utilization of image is consistently increasing in present era on multiple area e.g. domestic appliances, educational requirements, social networking, evidences in court cases, multimedia sharing, etc. There are various enterprise applications where image plays one of the most critical roles and the enterprise can suffer a collateral loss if such images or its contents are compromised by any means [1]. With an availability of modern image editing tools, it is now possible to re-create fake or forged image that area quite challenging to be identified as a real [2, 3, 4]. This phenomenon is called as image forgery [5]. Although, there are various types of attacks over an image, but splicing image is one of the frequently adopted practices to forge the image in social network or malicious spreading of fake propaganda. It merged two or more images or objects in one scene in such a way the final product looks unbelievably true. The fake image of Osama Bin Laden by Sun, UK Mail, and Telegraph is one live example of image splicing [6, 7]. Blitzer [8] has illustrated the underlying principle of image forensics where vivid description of many attacks can be studied. The core idea behind image forensic is to tamper and manipulate the image and give a new look into the image in such a way that it is not feasible to be identified as forged [9, 10]. While doing so an attacker has the equal chances of getting themselves caught. Hence, it is necessary to remove all the possible traces by the attacker in order to increase more imperceptibility towards splicing. This paper introduces on such algorithm in order to identify the position compromised by image splicing attack. Section 2 discusses about the recent research work towards image splicing accompanied by briefing of problem identification in Sect. 3. The proposed algorithm is briefed in Sect. 4 followed by algorithm description in Sect. 5. Comparative analysis of performance is discussed in brief in Sect. 6 and finally the contribution of the paper is highlighted in Sect. 7.

2 Related Work

Our prior work has reviewed about the effectiveness of the existing techniques towards enhancing detection of image forgeries [11, 12, 13]. Our prior techniques has introduced techniques towards copy-move attack and retouching attacks [14, 15]. In this part of the study, our interest orients towards image splicing attacks. The significant work towards splicing attack was carried out by Cozzolino et al. [16] by deploying expectation-minimization algorithm and feature extraction carried out by statistical approach. Adoption of real-world image was seen in the work of Zampoglou et al. [17] towards image splicing detection. Various optimization techniques e.g. singular vector decomposition, discrete cosine transformation, support vector machine etc. has been found to be adopted in various research work like that of Amerini et al. [18] and Moghaddasi et al. [19].

The focus of such techniques was mainly to find the exact position of the spliced region. Markov modeling is another technique for feature extraction to be used for identifying the underlying image structure by Su et al. [20]. Usage of Markov model for spliced region detection in colored image was seen in the work of Han et al. [21]. Local binary pattern is also used for feature extraction for identifying the splicing features found in the work carried out by Zhang et al. [22]. Usage of local descriptors for spliced region detection was also witnessed in work of Saleh et al. [23]. Similar type of scheme was also used by Zhao et al. [24] where the abnormal channels are explored from color channels considering chrominance feature. Usage of probability matrix along with the statistical approach was also found to be providing more information about the spliced region. Application of descriptive statistical measures e.g. skewness and kurtosis was seen in the work of Pan et al. [25] for spliced region detection. Adoption of cryptographic encryption was seen in the study of Niu et al. [26] in order to resist statistical attack. The authors have also used quaternary coding and chaos theory. The next section outlines the problems in existing system.

3 Problem Identification

A closer look into the existing techniques of detection of spliced region is found to use supervised learning techniques, complicated classification-based approaches, and principle component analysis. Unfortunately, such mechanism are computational expensive and are accompanied by degraded system reliability. Although there are various studies towards image splicing, but the direction is more into classification and less into capturing the unique correlated information among the image blocks. Usage of color filter array can solve the problem of forced blur mechanism while manually performing image splicing operation.

However, there are less number of attempts towards using statistical-based approach by considering the abnormalities in blurring effect which are either not removed while removal of traces by attacker or has significantly produced an artifacts. Therefore, the open issue of the research is to find cost-effective computational algorithm which has higher accuracy and computational less complex while identifying spliced region for a given forged image.

4 Algorithm for Spliced Region Detection (ASRD)

The proposed system targets to extract the spliced region for a given forged image. Figure 1 exhibits the methodology adopted for designing ASRD. The mechanism performs partitioning of multiple blocks for extracting significant attributes of every blocks and its correlation with neighbor blocks. This phenomenon significantly assists to investigate the tampering of any part of the image. The local attributes are extracted and the algorithm looks for artifacts that could be underlying in the image. As there are infinite pixels to be scanned and investigated hence statistical measures using probability theory are the most appropriate technique to do this job. The algorithm is also associated with a classification technique in order to recognize tampered and not tampered zone.
Fig. 1.

Schema of proposed ARDS

5 Algorithm Implementation

The algorithm is mainly responsible for identifying the spliced region for a given image. The algorithm takes the input of Bi j (block of image), Oi,j (original Resolution image), ρi,j (Gaussian Smoothening Function), σi,j (Noise), p (projection vector), Cm (Covariance matrix), vc/vp (vector representing complete corruption and partial corruption respectively), Th (Threshold), αclass1/αclass2 (classifier), etc., which after processing generates S (Spliced Region). The steps of the algorithms are as shown below:

Open image in new window

The algorithm initially converts the given image into grey scale B, which is then subjected to be partitioned by Bi,j blocks. The first step of the algorithm is to empirically represent the tampered portion of the image as shown in Line-2. The computation of ρi,j and Bi,j can be carried out using deconvolution technique. The consecutive step of the algorithm is to identify the type of tampering followed by classifying the formation of spliced region. For this, the algorithm considers an attribute η by integrating the dimensional parameter δ1 and standard deviation δ2. Empirically, it can be also represented as ηi,j = [δ1 δ2]T. The algorithm also computes the projection vector p as p = [ p1) p2)]T in order to yield better formulation of an attribute ηi,j using linear transformation where ηi,j = pT ηi,j. A covariance matrix Cm is formulated for both complete (Cc) and partial corrupted (Cp) image i.e. Cm = Cc−Cp. The study considers the binary possibility of two types of regions i.e. (i) completely corrupted or spliced region and (ii) non tampered region. So, the algorithm represents its projection vector with respect to this binary classification as shown in Line-3. Forming a logical condition of tampered region, the αclass1 is set for identification of both tampered and non-tampered spliced region (Line-4–7) using a particular threshold. The true positive and true negative parameters are then computed using αclass1 and αclass2 respectively. Depending upon the experimental value of αclass1 and αclass2, the threshold value can be fine tuned. The studies also performs check for cardinality of such spliced region (Line-8) and then highlight the spliced region using the binary classifier bin (Line-9). It will mean that as an outcome, the algorithm will make the entire image black with only the spliced region be explored as its natural or true color contents. Therefore, the proposed mechanism is able to identify spliced regions for number or any type of images corrupted by any degree of splicing operation.

6 Result Analysis

The analysis of the proposed study was carried out considering 1000 synthetic image dataset captured from SLR camera with varied range and resolution. Using existing image editing tool, they are also manipulated in order to obtain spliced image. Along with testing on synthetic dataset, the proposed system was also tested on standard datasets of Columbia Image Splicing database [27].

The uncompressed spliced image dataset is used from Columbia Image Splicing database as shown in Fig. 2. Figure 2 (a) and (b) Shows sample-1 image with two different probability map (of value 2 and 8) shows different pattern of peaks in histogram. It interprets that increase in probability also increase better detection rate of spliced region. Probability maps are normal maps of gray scale image in order to yield binary feature of classification. The binarization was carried out after fine-tuning the threshold Th that was selected based on higher value of true positive and negative value.
Fig. 2.

Visual outcomes of histogram analysis and probability in proposed system

The study outcome shown in Fig. 3 exhibits the visual outcomes of steps involved in processing. The spliced image is subjected for blocking operation that is further processed using binary classification technique illustrated in the algorithm to generate a binary image with tampered and non-tampered region. The elaborated discussion of the method used in implementation is as follows-The spliced region is varied from different ranges of sizes (smaller-bigger). The system then takes the spliced image as an input and multiple type of blocking operation is applied. The prime reason behind applying blocking operation is to ensure the detection performance. It also increases the feasibility of exploring the sensitive and critical area. Hence, it is recommended to apply smaller dimensions of the blocking operation in order to incorporate imperceptibility towards the spliced region. Moreover, maximized dimensions of the block partitioning can also be used for enhancing the difficulty level of an input spliced image. Once the finalization of the block partitioning operation is done then local level features are extracted. This operation is carried out using statistical-based approaches e.g. variance, mean, standard deviation, etc. Statistical-based features give more comprehensive information about the image and thereby over maximized benefits. Applying such forms of features extraction mechanism is quite deterministic in nature and therefore offers faster computation operations too. It is also cost effective in computational performance in contrast to any other statistical technique that uses inferential-based mechanism. The second advantage of this approach is its higher accuracy within inclusion of any recursive operation. Once the artifacts of the CFA are identified then the extraction of the outcome is carried out from the given image. Finally, the identification of the spliced region can be seen for the proposed system to retain its true color. For the purpose of an effective analysis, we compare outcome of proposed study ASRD with Ferrara et al. [28] and Han et al. [21] with respect to standard performance parameters of true positive, true negative, and accuracy. Figure 4 outlines the comparative performance analysis which takes the similar descriptive features discussed above in order to extract the spliced region. Both the technique i.e. proposed and existing system considers its own flow of mechanism involved in detecting forged region in order to finally obtain the numerical outcome with respect to accuracy in detection process.
Fig. 3.

Visual outcomes of proposed system

Fig. 4.

Comparative performance analysis

The outcome exhibits extensive identification of the spliced image as compared to Ferrara et al. [28] and Han et al. [21]. Though the difference is very marginal in terms of accuracy in detection factor, but proposed system has better computational capability in comparison to existing one. The complete algorithm processing time of ASRD was found to be 0.2765 s in core i7 machine while that of existing system was found to be approximately near to 1.2754 min. The memory complexity of the proposed system is also highly enhanced as it is free from any complex stochastic modeling like that of Han et al. [21] work. The performance of the proposed system with synthetic and standard dataset slightly differs in their outcome with 6.75%, which can be said to be within acceptable limits. Hence, the proposed study offers a robust and cost effective modeling for identification of regions within an image inflicted with image splicing operation. Apart from the accuracy, the response time of the proposed ASRD is found to be 75% of improvement as compared to the existing approaches of forged region detection in image processing.

7 Conclusion

The proposed study of ASRD has presented a technique that takes the forged image as an input in order to extract the precise region that has been maliciously tampered or corrupted with image splicing attack. The complete ideology of the proposed study is about uniformity among the neighborhood pixels with each in an original image. This uniformity is broken during image splicing in such a way that it is very difficult to perform identification of traces based on pixels. Hence, the proposed system presents a mechanism that performs partitioning of the blocks in order to obtain better granularity in the investigational findings. The study doesn’t use any complex or iterative algorithms of optimization what can be seen in abundant in existing research techniques. This is where the proposed ASRD makes a different by introducing a very simple and cost effective algorithm for identifying the spliced region.

References

  1. 1.
    Shih, F.Y.: Multimedia Security: Watermarking, Steganography, and Forensics. CRC Press, Boca Raton (2012)CrossRefGoogle Scholar
  2. 2.
    Julliand, T., Nozick, V., Talbot, H.: Image noise and digital image forensics. In: Shi, Y.-Q., Kim, H.J., Pérez-González, F., Echizen, I. (eds.) IWDW 2015. LNCS, vol. 9569, pp. 3–17. Springer, Cham (2016). doi:10.1007/978-3-319-31960-5_1 CrossRefGoogle Scholar
  3. 3.
    Ding, F., Dong, W., Zhu, G., Shi, Y.-Q.: An advanced texture analysis method for image sharpening detection. In: Shi, Y.-Q., Kim, H.J., Pérez-González, F., Echizen, I. (eds.) IWDW 2015. LNCS, vol. 9569, pp. 72–82. Springer, Cham (2016). doi:10.1007/978-3-319-31960-5_7 CrossRefGoogle Scholar
  4. 4.
    Choi, C.-H., Lee, M.-J., Hyun, D.-K., Lee, H.-K.: Forged region detection for scanned images. Springer-Comput. Sci. Converg. 114, 687–694 (2011)CrossRefGoogle Scholar
  5. 5.
    Malviya, P., Naskar, R.: Digital forensic technique for double compression based JPEG image forgery detection. Springer-Inf. Syst. Secur. 8880, 437–447 (2014)Google Scholar
  6. 6.
    Smith, S.: iMediaEthics’ Top 10 Fake and Doctored Photo Stories. An online article of iMediaEthics 2016. http://www.imediaethics.org/imediaethics-top-10-fake-and-doctored-photo-stories/. Accessed 20 Oct
  7. 7.
    Vamosi, R.: Researcher: Bin Laden’s beard is real, video is not. An online article of CNET. https://www.cnet.com/news/researcher-bin-ladens-beard-is-real-video-is-not/. Accessed 20 Oct 2016
  8. 8.
    Blitzer, H.L., Stein-Ferguson, K., Huang, J.: Understanding Forensic Digital Imaging. Academic Press, Cambridge (2010)Google Scholar
  9. 9.
    Stamm, M.C., Liu, K.J.R.: Forensic detection image manipulation using statistical intrinsic fingerprints. IEEE Trans. Inf. Forensics Secur. 5(3), 492–506 (2010)CrossRefGoogle Scholar
  10. 10.
    Weisi, L., Tao, D., Kacprzyk, J., Li, Z., Izquierdo, E.: Haohong Wang, Multimedia Analysis, Processing and Communications. Springer Science & Business Media, New York (2011)Google Scholar
  11. 11.
    Sundarm, A.M., Nandini, C.: Copy-move forgery detection- a survey. In: ICACCN-International Conference on Advanced Computing, Communication Networks, Chandigarh, 02–03 June 2011Google Scholar
  12. 12.
    Sundarm, A.M., Nandini, C.: Investigational study of image forensic applications, techniques and research directions. Int. J. Emerg. Technol. Adv. Eng. Certif. J. 4(8), 1–9 (2014). https://www.ijetae.com. ISSN 2250-2459, ISO 9001:2008Google Scholar
  13. 13.
    Sundarm, A.M., Nandini, C.: Image retouching and it’s detection-a survey. In: NCGCT-First National Conference on Green Computing Technologies, DSATM, Bangalore, 07 March 2015Google Scholar
  14. 14.
    Sundarm, A.M., Nandini, C.: Feature based image authentication using symmetric surround saliency mapping in image forensics. Int. J. Comput. Appl. 104(13), 1–9 (2014)Google Scholar
  15. 15.
    Sundarm, A.M., Nandini, C.: CBFD: coherence based forgery detection technique in image forensics analysis. In: IEEE-ICERECT-2015-International Conference on Emerging Research in Electronics, Computer Science and Technology, 17–19 December 2015Google Scholar
  16. 16.
    Cozzolino, D., Poggi, G., Verdoliva, L.: Splicebuster: a new blind image splicing detector. In: IEEE International Workshop on Information Forensics and Security (2015)Google Scholar
  17. 17.
    Zampoglou, M., Papadopoulos, S., Kompatsiaris, Y.: Detecting image splicing in the wild (web). In: IEEE International Conference on Multimedia and Expo Workshop, pp. 1–6 (2015)Google Scholar
  18. 18.
    Amerini, I., Becarelli, R., Caldelli, R., Mastio, A.D.: Splicing forgeries localization through the use of first digit features. In: IEEE International Workshop on Information Forensics and Security (2014)Google Scholar
  19. 19.
    Moghaddasi, Z., Jalab, H.A., Md Noor, R.: SVD-based image splicing detection. In: International Conference on Information Technology and Multimedia (2014)Google Scholar
  20. 20.
    Su, B., Yuan, Q., Wang, S., Zhao, C., Li, S.: Enhanced state selection Markov model for image splicing detection. Springer-EURASIP J. Wirel. Commun. Netw. 2014, 1–10 (2014)CrossRefGoogle Scholar
  21. 21.
    Han, J.G., Park, T.H., Moon, Y.H., Eom, K.: Efficient Markov feature extraction method for image splicing detection using maximization and threshold expansion. J. Electron. Imaging 25(2), 023031 (2016)CrossRefGoogle Scholar
  22. 22.
    Zhang, Y., Zhao, C., Pi, Y., Li, S.: Revealing image splicing forgery using local binary patterns of DCT coefficients. In: Liang, Q., et al. (eds.) Springer Journals of Communications, Signal Processing, and Systems. LNEE, pp. 181–189. Springer, New York (2012). doi:10.1007/978-1-4614-5803-6_19 CrossRefGoogle Scholar
  23. 23.
    Saleh, S.Q., Hussain, M., Muhammad, G., Bebis, G.: Evaluation of image forgery detection using multi-scale weber local descriptors. In: Bebis, G., et al. (eds.) ISVC 2013. LNCS, vol. 8034, pp. 416–424. Springer, Heidelberg (2013). doi:10.1007/978-3-642-41939-3_40 CrossRefGoogle Scholar
  24. 24.
    Zhao, X., Li, S., Wang, S., Li, J., Yang, K.: Optimal chroma-like channel design for passive color image splicing detection. Springer-EURASIP J. Adv. Signal Process. 2012, 240 (2012)CrossRefGoogle Scholar
  25. 25.
    Pan, X., Zhang, X., Lyu, S.: Exposing image splicing with inconsistent local noise variances. In: IEEE International Conference on Computational Photography, pp. 1–10 (2012)Google Scholar
  26. 26.
    Niu, H., Zhou, C., Wang, B., Zheng, X., Zhou, S.: Splicing model and hyper-chaotic system for image encryption. J. Electr. Eng. 67(2), 78–86 (2016)Google Scholar
  27. 27.
    Columbia Image Splicing Detection Evaluation Dataset. http://www.ee.columbia.edu/ln/dvmm/downloads/AuthSplicedDataSet/AuthSplicedDataSet.htm, Retrieved 06th February 2017
  28. 28.
    Ferrara, P., Bianchi, T., De Rosa, A., Piva, A.: Image forgery localization via fine-grained analysis of CFA artifacts. IEEE Trans. Inf. Forensics Secur. 7(5), 1566–1577 (2012)CrossRefGoogle Scholar

Copyright information

© Springer International Publishing AG 2017

Authors and Affiliations

  1. 1.School of Computing and Information TechnologyREVA UniversityBangaloreIndia
  2. 2.Department of Computer Science and EngineeringDayanand Sagar Academy of Technology and ManagementBangaloreIndia

Personalised recommendations