Available online at www.sciencedirect.com ScienceDirect Procedia Computer Science 85 (2016 ) 206 212 International Conference on Computational Modeling and Security (CMS 2016) Survey On Keypoint Based Copy-move Forgery Detection Methods On Image Devanshi Chauhan a *, Dipali Kasat b, Sanjeev Jain c, Vilas Thakare d a,b Computer Engineering, Sarvajanik College of Engineering & Technology, Surat 395001,India c Computer Engineering, MITS, Gwalior 474007, India d Computer Science & Engineering, Amravati University, Amravati 444602, India Abstract One of the problem in image forensics is to check the authenticity of image. This can be very important task when images are used as an evidence which cause change in judgment like, for example in a court of law. Image is forged by using different techniques but in that most common technique is copy-move forgery. Copy-move forgery is created by copying the region from a particular image and pasting that region on same image to mislead the user. This type of forgery is done using availability of new sophisticated software and applications. This type of forgery is also done in video. In this paper we survey on different keypoint based copy-move forgery detection methods with different parameters. 2015 The Authors. Published by Elsevier B.V. 2016 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of organizing committee of the 2016 International Conference on Computational Modeling and Peer-review Security (CMS under 2016). responsibility of the Organizing Committee of CMS 2016 Keywords: image forgery; image forgery detection; copy-move forgery; copy-move forgery detection; image splicing; image retouching; copymove forgery methods; keypoint based methods 1. Introduction The main force behind the digital image forensics is image forgery. Now a days video and images hold high importance because they have become a main source of information. Video/images are very useful in various field like medical imaging, digital forensics, intelligence, sports, scientific publications, journalism, etc, Due to various * Corresponding author. Tel.:+91-814-112-2123 E-mail address:chauhandevanshi7893@gmail.com 1877-0509 2016 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of the Organizing Committee of CMS 2016 doi:10.1016/j.procs.2016.05.213
Devanshi Chauhan et al. / Procedia Computer Science 85 ( 2016 ) 206 212 207 software like Photoshop, Coral draw for image, Premiere, Vegas for videos and android applications are also there like photo hacker copy & paste, it is sometimes difficult to distinguish between original image/video and the forged image/video. For example, if crime scene was recorded and forged by criminal for malicious purpose or defense purpose, then that image/video can t be used as an evidence in court of law. In all over the world many persons are saw the news. In that they saw the video/images. But news makers also can t prove that video played or image showed by news channel is trustworthy. And also the person who viewed that video or image may not be sure that it is real or forged. The authenticity of video/image plays a very important role in the aspects of sensitive cases where the video is produced as a witness in court of law, so even a small modification is not acceptable in that case. However other cases are there in which modification is allowed in image or videos such as in shooting. After completion of shooting journalist may need to edit video/images. After that journalist broadcast that video/images on channe l. As shown in Fig. 1 image forgery is divided into two categories: Active approach and Passive approach. In active approach, the concept of digital watermarking and digital signature or the combination of both is used. In active approach detector has prior information about image like from which camera that image was taken. In passive methods tampering is detected. In that detector has no prior information about digital signature and digital watermarking. If there is no information about image that from which camera image has been taken.,it is called blind image. There are many cameras available which provide digital watermark or signature like Epson PhotoPC 700/750Z, 800/800Z, 3000Z and Kodak DC290 4. Passive approach is divided into three categories: 1) Copy-move forgery 2) Image Splicing 3) Image Retouching 1.1. Copy-move forgery The copy-move forgery technique is used to hide some sensitive or important information. As shown in Fig. 3 army truck was hidden by using tree s portion. In that tree s portion copy and paste on that truck to hide. No one can detect easily that image is original or forged. Basic process of copy-move forgery detection is shown in Fig. 2. Copy-move forgery detection is either keypoint based or block based. In block based methods image is divided into the rectangular regions. And keypoint based methods extract feature point only on particular regions from an image without any subdivisions of an image. In both case preprocessing of the images is done such as conversion of image into gray scale, etc. Next is feature extraction, blockbased methods subdivide the image in rectangular regions. For every such region, a feature vector is computed. Similar feature vectors are subsequently matched. And in keypoint based method, it extract feature point using different methods like SIFT, SURF etc without any image subdivision. The Feature points are matched with each other using different approaches like clustering, Euclidean distance. A forgery shall be reported if matching features are found 3. After that filtering is applied to remove spurious matches. And the last step post-processing is used to analyze filtered result for forgery detection 3. Mostly keypoint based methods are used to extract features. Because in block based methods image is divided into similar s. Then blocks are processed and then inter-related to identify any possibility of forgery. So major drawback of this method is it requires exponentially large number of comparisons as image size increases 5.
208 Devanshi Chauhan et al. / Procedia Computer Science 85 ( 2016 ) 206 212 1.2. Image splicing Image splicing is another kind of forensics method. In image splicing a single image is created by combination of two or more images. As shown in Fig. 4(a) and (b) are two original images. Using both images new forged image generated in Fig. 4(c) is called splicing image. Image splicing is a challenging problem because in that joining regions are there. It is hard to extract perfect shape wise object of image to make new image. Image splicing is categorized in two types: Region based and boundary based. Image splicing is a challenging problem because in that joining regions are there. It is hard to extract perfect shape wise object of image to make new image. Image splicing is categorized in two types: Region based and boundary based 1. 1.3. Image retouching Image retouching is one type of forensic method. In image retouching slight changes are made in image such as change in weather, colour, make blurred background, etc. As shown in Fig. 5 right one is replaced with left one. Other example of image retouching is shown in Fig. 6.And this type of forgery undergo geometric transformation like scaling, rotation, stretching, etc to create a new forged image. As shown n Fig. 7 airplane is rotated and then pasted in the same image to generate a new misleading image. In this paper we survey only on different copy-move forgery detection methods using keypoint based methods on image. Fig. 1 classification of image forgery Fig. 2 basic process for detection of copy move forgery 3,5 Fig. 3 (a) original image (b) forged image 1,10
Devanshi Chauhan et al. / Procedia Computer Science 85 ( 2016 ) 206 212 209 Fig. 4 (a) original image1 (b) original image2 (c) spliced image 1,2 Fig. 5 right one image is replaced by left one 1 Fig. 6 (a) original image (b) color change (c) weather change (d) background blur 1 Fig. 7 example of image retouching 2 2. Copy-move forgery methods In image forgery copy-move forgery detection is most important issue. As shown in Fig. 8 copy-move forgery detection hierarchy is mainly divided into three categories: Block based, keypoint based and brute force based. Copy-move forgery detection hierarchy is shown in Fig. 8. Brute force is detected by using exhaustive search. In brute force image with circularly shifted versions are use to examine matching segments. Its computational complexity is very high to make such type of comparisons. Autocorrelation determine changes of location. In block based method images are first divided into square blocks or circular blocks. Block based method is further categorized into two types: Spatial Domain and Transform Domain. Spatial Domain directly deal with pixels. It compare blocks with its pixel. Other type is transform domain in which different transformation methods are use to detect copy move forgery. It is again further categorized into two types: Transform only and post-transform supported AI/statistical processing. Transform only method use DWT(Discrete Wavelet Transform), DCT(Discrete Cosine Transform), etc. Post-transform supported AI/statistical processing with DWT, DCT techniques. Last type of copy- move forgery is keypoint based methods in which SIFT(Scale Invariant Feature Transform), MIFT(Mirror reflection Invariant Feature Transform) and SURF (Speed up Robust Feature) are used to extract feature points from images. In next section we survey on different copy-move forgery detection techniques using keypoint based method.
210 Devanshi Chauhan et al. / Procedia Computer Science 85 ( 2016 ) 206 212 Fig. 8 classification of copy-move forgery 3. Comparison of keypoint based copy move forgery techniques In image forgery copy-move forgery detection is most important issue. SIFT, SURF, MIFT are keypoint based methods which are use to extract feature points. Table. 1 exhibits the comparison of different copy-move forgery detection techniques using keypoint based methods. Here the five parameters are used. 1) Method used for feature extraction, 2) Strategy used for feature matching, 3) pre-processing method, 4) block pattern and 5) performance. The method used for feature extraction indicate which method is use to extract feature points from image. After that strategy used for feature matching indicate which strategy is used for matching feature. Pre-processing method indicate which process is used at the starting of the detection process. Parameter block pattern can detect the presence of block strategy in detection process. In case of presence indicate which type of blocks used. Last is performance, detect the performance of keypoint based copy-move forgery methods. Discussion: From the comparison shown in Table. 1 it is clear that SIFT, SURF, and MIFT are the three major techniques found for detecting copy-move forgery in images. SURF use normal pixel s and can detect geometrical transformation like scaling, rotation or transformation performed after forgery. SURF use block based method which has high computational complexity. MIFT use small s. So it is an efficient technique to detect forge in close and neighboring regions. But MIFT suffers from high computational complexity problem. SIFT is an efficient technique and can detect forgery in a single or multiple regions of an image. Also it is considered to give good detection results in case of both plain copy-move forgery and geometric transformation like scaling, rotation, translation. But SIFT is invariant to rotation, scaling and affine transformation. And SIFT give high computational efficiency compared to SURF. But SIFT accuracy is low compared to SURF.
Devanshi Chauhan et al. / Procedia Computer Science 85 ( 2016 ) 206 212 211 Reference No. [5] Table 1. Comparison of different keypoint based copy-move forgery detection methods. Method used for feature extraction Strategy for feature matching Coefficient map and threshold [6] Transform Matrix [7] Euclidean distance Pre-processing method DWT and segmentation Detection Region/s Multiple Block pattern Irregular and non-overlapping blocks Segmentation Single Small independent patches Segmentation Multiple Normal pixel [8] BFSN clustering No Multiple Normal pixel [9] Pattern entropy based near duplicate SIFT detection [10] Euclidian distance and threshold [11] Hierarchical clustering [12] Use descriptor vector [13] MIFT RANSAC and hysteresis thresholding [14] SURF Use descriptor vector [15] SIFT and SURF Euclidian distance Divide region using mean shift clustering Performance Give much better result under various conditions like plain copy move and different transform Expectation Maximization based algorithm give better performance using SIFT 1)Give minimum time complexity, 2)Given method is robust, simple, efficient, and invariant to scale and rotation of pasted object. 1)Method is effective when more than one group of tampered region exist in tampered image. 2)Robustness against rotation and scaling Single Square region Improve the accuracy of forgery detection, Low computational complexity No Single Normal pixel No Multiple Normal pixel DWT Single Normal pixel Apply geometric constraints (slope, location) Due to strong stability of SIFT give good performance in different kind of post image processing Extract higher number of feature points and Detect multiple copied region Computational efficiency increase and detect even if copied part is rotated/scaled then pasted Single Small block Detect forged region with high accuracy and robustness DyWT and DWT Single Normal pixel No Single Normal pixel Detect scaling and rotated object, reduce computational complexity SIFT and SURT give fast and robust performance with respect to geometrical transformation
212 Devanshi Chauhan et al. / Procedia Computer Science 85 ( 2016 ) 206 212 4. Conclusion In this paper we survey on different copy-move forgery detection techniques using keypoint based methods on forged image. This survey will help to researchers to improve detection with new ideas and new challenges. We have identified that some methods are not responsive for geometric transformation such as scaling and rotating. Also we have noted some methods which give accuracy but has high computational complexity. References 1. Qazi, Tanzeela. "Survey on blind image forgery detection." IET, 2013. 2. Ali Qureshi, M., and M. Deriche. "A review on copy move image forgery detection techniques." IEEE, 2014. 3. Christlein, Vincent, et al. "An evaluation of popular copy-move forgery detection approaches." IET, 2012. 4. Wahab, Ainuddin Wahid Abdul, et al. "Passive video forgery detection techniques: A survey." IEEE, 2014. 5. Pun, Chi-Man, Xiao-Chen Yuan, and Xiu-Li Bi. "Image Forgery Detection Using Adaptive Over-Segmentation and Feature Points Matching." IEEE, 2015. 6. Li, Jian. "Segmentation-based Image Copy-move Forgery Detection Scheme." IET, 2015. 7. Sudhakar, K., V. M. Sandeep, and Santosh Kulkarni. "Speeding-up SIFT based copy move forgery detection using level set approach." IEEE, 2014. 8. Liu, Lu, et al. "Improved SIFT-Based Copy-Move Detection Using BFSN Clustering and CFA Features." IEEE, 2014. 9. Keum, Ji-Soo, Hyon-Soo Lee, and Manabu Hagiwara. "Mean shift-based SIFT keypoint filtering for region-of-interest determination." IEEE, 2012. 10. Huang, Hailing, Weiqiang Guo, and Yu Zhang. "Detection of copy-move forgery in digital images using SIFT algorithm." IEEE, 2008. 11. Amerini, Irene, et al. "A sift-based forensic method for copy move attack detection and transformation recovery." IET, 2011. 12. Hashmi, Mohammad Farukh, Aaditya R. Hambarde, and Avinash G. Keskar. "Copy move forgery detection using DWT and SIFT features." IEEE, 2013. 13. Jaberi, Maryam, et al. "Improving the detection and localization of duplicated regions in copy-move image forgery." IEEE, 2013. 14. Hashmi, Mohammad Farukh, Vishal Anand, and Avinash G. Keskar. "A copy-move image forgery detection based on speeded up robust feature transform and Wavelet Transforms." IEEE, 2014. 15. Pandey, Ramesh Chand, et al. "Fast and robust passive copy-move forgery detection using SURF and SIFT image features." IEEE, 2014.