You are on page 1of 7

IPASJ International Journal of Computer Science (IIJCS)

A Publisher for Research Motivation ........

Volume 2, Issue 1, January 2014

Web Site: http://www.ipasj.org/IIJCS/IIJCS.htm Email: editoriijcs@ipasj.org ISSN 2321-5992

Detection of Copy-Move Forgery using Haarlike features


Mehdi Ghorbani1, Mohammad Firouzmand2
1

Young researcher and Elites Club, Miyaneh Branch Islamic Azad University, Miyaneh, Iran
2

Iranian Research Organization for Science & Technology, Tehran, Iran

ABSTRACT
Copy-move forgery is a specific type of image tampering where a part of the image is copied and pasted on another part generally to conceal unwanted portions of the image. Hence, the goal in detection of copy-move forgeries is to detect image areas that are same or extremely similar. This paper describes a method for robust and accurate copy-move forgery detection by employing Haar-like features on the integral image. By applying Haar-like filters to nn overlapping blocks of the integral image instead of input image, time needed for the detection process is reduced. Experimental results show that the proposed scheme accurately detects such specific image manipulations although the copied region are rotated and its complexity time is very small compared to approaches base on overlap blocking.

Keywords: Copy-Move Forgery, Duplicated Image Regions, integral image, Haar-like features

1. INTRODUCTION
It was a very difficult task in the old time without digital cameras and computers to create a good splicing photograph, which requires sophisticated skill of darkroom masking. Due to rapid advances in powerful image processing software, digital images are easy to manipulate and modify. This makes it more difficult for humans to check the authenticity of a digital image. Nowadays, modifying the content of digital images becomes much easier with the help of sophisticated software such as Adobe Photoshop. Region duplication or region cloning is a very common practice of image tampering, where a continuous portion of pixels in an image is pasted to a different location to conceal undesirable objects or contents in the original image. Figure 1, for example, shows a copy-move forgery where a part of the background is copied and moved to hide the person appearing in the original image. It can be observed from this example that such image manipulations may not leave obvious evidence of tampering. The necessity of algorithms for efficiently verifying the integrity of images cannot, therefore, is overemphasized in this digital era.

(a) (b) Figure 1 Example of Copy-Move forgery :(a) Original image, (b) Tampered image Recently, numerous techniques for image integrity verifications have been proposed. Some techniques employ watermarking schemes [1]-[3] to authenticate an image as well as determine its integrity. The drawback with schemes based on watermarking is that the watermark must be embedded right during the image formation to avoid the

Volume 2 Issue 1 January 2014

Page 1

IPASJ International Journal of Computer Science (IIJCS)


A Publisher for Research Motivation ........

Volume 2, Issue 1, January 2014

Web Site: http://www.ipasj.org/IIJCS/IIJCS.htm Email: editoriijcs@ipasj.org ISSN 2321-5992

possibility of watermarking an already forged image. This is practically difficult as most digital cameras and other image acquisition devices do not have instantaneous watermarking facilities. There are also various techniques that detect image tampering in absence of watermarks and signatures. For instance several methods have been proposed to detect region duplication for the purpose of image forensics [4]-[6]. These methods are based on finding pixel blocks that are exact copies of each other in an image. Such methods are most effective for the detection of region copy-move, where a region of pixels is pasted without any change to another location in the image. Also each of the schemes mentioned above commands meaningful efficiency only in specific kinds of tampering and effort to reduce of computation. To further reduce the amount of computation, this paper proposes new Haar-like features based approach. The rest of this paper is organized as follows. In Section 2, we present our proposed method in details and demonstrate its steps. Section 3 presents results of the method when applied to real forged images. Section 4 concludes the paper.

2. THE PROPOSED METHOD


Given an arbitrary image, the goal of forgery detection is to determine whether or not there are any duplicate regions in the images and, if present, return the image location and extent of each region. An obvious approach is to exhaustively compare every possible pair of regions. However, such an approach is exponentially complex. Block matching appears to be a more efficient which slides a nn window over an MN image pixel by pixel resulting in k = (M-n+1)(N-n+1) blocks. Each block is column-wisely reshaped into n2 long row vector, otherwise known as feature vector, and inserted into a kn2 feature matrix [7]. Since in methods based on overlap blocking, time complexity is associated with the number and size of the blocks, therefore recently efforts have been made in reducing blocks' number and size [8], [9]. With respect to haar-like filters and integral image, Viola and M. Jones [10], [11] proposed a new face detection method. They applied rectangle black and white shapes to find the face in the image that reduces the processing time substantially. All rectangles had different weights and scales to reflect the characteristics of a human face. Now we can use filters in detection of digital image copy-move forgery. The steps of forgery Detection in proposed method are as shown in Figure 2.

Figure 2 Forgery detection steps Step I

Volume 2 Issue 1 January 2014

Page 2

IPASJ International Journal of Computer Science (IIJCS)


A Publisher for Research Motivation ........

Volume 2, Issue 1, January 2014

Web Site: http://www.ipasj.org/IIJCS/IIJCS.htm Email: editoriijcs@ipasj.org ISSN 2321-5992

Read the image selected by the user as input. If the input image is not a gray scale image then convert it into a gray scale image. Step II Integral image: The integral image is implemented by summing over the pixel values line by line, starting from the top left corner of the image. Each pixel is composed of the sum of preceding pixels in the row of the input image and the integral value of the pixel in the line above. The integral image is also called a summed area table (SAT) as is contains the sum of the current pixel plus all pixels above and left of the current pixel. This can be written as in (1). (1) Finding the sum in an arbitrary rectangular area of the SAT is now a simple calculation with constant computation time, as shown in (2) and illustrated in Figure 3. (2)

Figure 3 Illustration of the area calculation in the summed area table Step III Haar-like features: Haar-like features are rectangular features that can indicate specific characteristics in an image. The idea behind Haar-like features is to recognize objects or features based on the value of simple features, instead of pixel values directly. The Haar-like features have the advantage of very fast computation, because it depends only on the sum of pixels within a rectangle instead of every pixel value. Using an integral image for calculating the sum, one rectangle can be computed with only four references, independent of the size of the feature, as explained in step II. The simplest Haar-like feature is a two-rectangle feature. The value of this is calculated as the difference between the sums of the pixels within the two rectangles. This will indicate characteristics like edges or borders between light and dark regions. Three-rectangle features indicate for instance a dark line or a dark thin area lying between light regions, depending on the size of the middle rectangle. Four-rectangle features compute the difference between diagonal pairs of rectangles, and so on. Using the integral image for computing the sum will allow a two-rectangle feature to be calculated with six references to the integral image, a three-rectangle feature with eight references, a four-rectangle feature with nine references and etc. Instance of Haar-like features is shown in Figure 4.

Figure 4 Original haars and their rotated filter set Step IV Comparison of the results of the selected filters: as seen in Figure 2, the sum of pixels is computed by integral image and nodes of rectangles are obtained. Using the nodes, haar-like features are calculated by using the filters and the result is compared with other filters result. We can by applying different filters obtain specific characteristics in an image with different angles.

Volume 2 Issue 1 January 2014

Page 3

IPASJ International Journal of Computer Science (IIJCS)


A Publisher for Research Motivation ........

Volume 2, Issue 1, January 2014

Web Site: http://www.ipasj.org/IIJCS/IIJCS.htm Email: editoriijcs@ipasj.org ISSN 2321-5992

For example, in forgery image that duplicate region is rotate with 90 angles, detection or forgery will be discoverable by applying rotation filter with 90 angles. In Figure 5 is shown block which on its rotation filters is applied. However, in practice, these filters applied to the integral image and not need to sum of all block elements, we need only to element four of integral image.

Figure 5 The block rotated and applying of rotation filter Step V Calculation of the shift vectors: Two copied and then moved areas will yield some pair of identical features. This feature will yield same shift vector. This shift vector will be checked for a particular number of neighbors. Same shift vector will be showing the duplicated region as shown in Figure 6.

Figure 6 Neighbor Shift Matching For blocks that where vector a count is kept of the number of times it examined, the corresponding pair of positions in highlighted. each pair of adjacent is same, compute associated normalized shift vector and are associated block positions and for each shift is seen. The shift vectors with are the image are found and the nn blocks they represent are

3. EXPERIMENTAL RESULTS
This section presents the results of the described detection method for a variety of real images. The haar-like feature responses are found by matching the haar-like features with the input picture using the integral image. According to Figure 2, the input image is converted to gray scale and then to reduce its time complexity the integral image is obtained. Due to the angle of rotation, specific filters are used. For images of the repetitive area without rotating them one simple filter or two is enough. In cases where more than one filter is used, the results of all filters must be performed on each block and due to the shift vector; blocks are selected as the duplicate area. We consider block size pixels, and assume that the minimum size of the repetitive area to be . So the ideal frequency of shift vector is set . However, in anticipation of manipulation any frequency greater than , passes for duplication. In addition, increasing the block size, the discovery of forgery is carried out accurately as long as smaller than repetitive area. Initially, it is not known what the size of the duplicated region is. So it becomes expedient that to detect even small size patches, much smaller block size be selected. However, with every small block size, chances of catching even small similar regions become substantial. As a result, there may be various false matches. Figure 7(b) shows tampered images where the region is duplicated. In Figure 7(c), we show the results of our method performed on forgery images. As can be seen, when the repetitive region is without rotation, detection of copy-move forgery can be done more quickly and accurately than other relative algorithms.

Volume 2 Issue 1 January 2014

Page 4

IPASJ International Journal of Computer Science (IIJCS)


A Publisher for Research Motivation ........

Volume 2, Issue 1, January 2014

Web Site: http://www.ipasj.org/IIJCS/IIJCS.htm Email: editoriijcs@ipasj.org ISSN 2321-5992

(a) (b) (c) Figure 7 (a) Original image, (b) image with duplication, (c) The result of detecting the duplication With the aid of just two simple filters the duplicate region without rotation and rotation of 90, 180, 270 angles can be detected with greater speed and accuracy than the previous methods. For detection of other rotations, we can use rotating filters. Figure 8 shows the result of rotating the images.

(a) (b) (c) Figure 8 (a) Original image, (b) image with duplication, (c) The result of detecting the duplication In order to detect duplicate region with rotation, rotation filters are used. The results of applying these filters adapted and considered to regions as candidate. Then with normalized vector shifts, repeated regions are identified. Figure 9 shows the results of applying rotation filters on manipulate images with different rotations.

4. CONCLUSION
In this paper, an improved algorithm based on Haar-like features on the integral image to detect digital images copymove forgery was proposed. Integral image was used to reduce the time complexity in which instead of comparing blocks elements; it is enough to compare the number which is achieved from applying filters. Also for detecting duplicate regions with rotation angle rotation filters with different angles was used. Experimental results show that time complexity is reduced compared with the existing related algorithms based on overlap blocking. Furthermore, detection of rotated regions is possible. In future, we intend to search for features that are invariant to manipulations such as rescaling.

Volume 2 Issue 1 January 2014

Page 5

IPASJ International Journal of Computer Science (IIJCS)


A Publisher for Research Motivation ........

Volume 2, Issue 1, January 2014

Web Site: http://www.ipasj.org/IIJCS/IIJCS.htm Email: editoriijcs@ipasj.org ISSN 2321-5992

Figure 9 Result of algorithm on tampered images with different rotation angles

References
[1] C. T. Hsieh, Y. K. Wu, Geometric Invariant Semi-fragile Image Watermarking Using Real Symmetric Matrix, Journal of WSEAS Transaction on Signal Processing, Vol. 2, Issue 5, pp. 612-618, 2006. [2] P. Meerwald, and A. Uhl, A Survey of Wavelet-Domain Watermarking Algorithms, in Proceedings of SPIE, Electronic Imaging, Journal of Security and Watermarking of Multimedia Contents, Vol. 4314, pp. 505-516, 2001. [3] F. Hartung, and M. Kutter, Multimedia Watermarking Techniques, Journal of in Proceedings of the IEEE, Vol. 87, No. 7, pp. 1079-1107, 1999. [4] G. Li, Q. Wu, D. Tu, and S. Sun, A Sorted Neighborhood Approach for Detecting Duplicated Regions in Image Forgeries based on DWT and SVD, in Proceedings of IEEE International Conference on Multimedia and Expo, Beijing China, pp. 1750-1753, 2007. [5] S. Khan, A. Kulkarni, Reduced Time Complexity for Detection of Copy-Move Forgery Using Discrete Wavelet Transform, in International Journal of Computer Applications Volume 6. No.7, 2010. [6] P. Yadav, Y. Rathore, Detection of Copy-Move Forgery of Images Using Discrete Wavelet Transform, International Journal on Computer Science and Engineering, Vol. 4 No. 04, 2012. [7] J. Fridrich, D. Soukal, and J. Lukas, Detection of Copy-Move Forgery in Digital Images, in Proceedings of Digital Forensic Research Workshop, 2003. [8] M. Zimba, S. Xingming , DWT-PCA (EVD) Based Copy-move Image Forgery Detection, in International Journal of Digital Content Technology and its Applications, PP. 251-258, 2011. [9] M. Ghorbani, M. Firouzmand and A. Faraahi, DWT-DCT (QCD) Based Copy-Move Image Forgery Detection, in 18th International Conference on Systems, Signals and Image Processing (IWSSIP 2011) , 2011. [10] P.Viola and M. Jones, Rapid object detection using a boosted cascade of simple features, Journal of Proceedings of IEEE Conf. on Computer Vision and Pattern Recognition (CVPR 2001), vol.1, pp. 511-518, 2001. [11] P.Viola and M. Jones, Robust real-time object detection, International Journal of Computer Vision, vol. 57, no. 2, pp. 137-154, 2004.

Volume 2 Issue 1 January 2014

Page 6

IPASJ International Journal of Computer Science (IIJCS)


A Publisher for Research Motivation ........

Volume 2, Issue 1, January 2014


AUTHOR

Web Site: http://www.ipasj.org/IIJCS/IIJCS.htm Email: editoriijcs@ipasj.org ISSN 2321-5992

Mehdi Ghorbani received the B.S. degrees in Computer Science from Payam Noor University of Miyaneh Center in 2003-2007 and M.S. degrees in Computer Engineering from Payam Noor University of Rey Center in 2007-2010. He is currently remembered in young researcher and Elites Club, Miyaneh Branch Islamic Azad University and research in context of image processing.

Mohammad Firouzmand received his M.S.degree in Biomedical Engineering in Sharif Univ. in 1992 and his Ph.D.degree in signal processing in Institute National Polytechnique de Grenoble (INPG), France, in 2007. He is currently an Associate Professor in Iranian Research Organization for Science and technology (IROST). His research interests include Bio-instruments, signal processing, image and audio signal processing, and he has published several research papers in these areas.

Volume 2 Issue 1 January 2014

Page 7

You might also like