A Review on Detection Strategies of Digital Image Tamper

Marwan Saad Kadhim

AL-Mustansiriya University – Iraq

Marwan_saad_k@yahoo.com

ABSTRACT

Today our world is driven by information and technological breakthroughs which is characterized mainly by spreading of digital images. The most common form of transferring information is through magazines, newspapers, scientific journals, or the internet.

This huge use of images technology has been accompanied by an evolution in editing tools of image processing which makes modifying and editing images very simple. Nowadays, the circulation of such forgery images, which distort the truth, has become common, intentionally or unintentionally. There are many types of digital image forgery but the most important and popular type is Copy-Move. It is easy to implement and difficult to detect.

Because the copied and pasted region is from the same image and the post-processing manipulations are performed on the entire image, the characteristics of the copied area (s), are compatible with that image. Therefore, the usual approaches of detecting incompatibilities that use statistical measurements to compare several portions of the image are not useful for Copy-Move forgery detection. Furthermore, several types of transforms can be performed to make the tampered image with more realistic.

مراجعة لاستراتيجيات اكتشاف العبث الرّقميّ للصور

مروان سعد كاظم الجامعة المستنصريّة/ العراق

الملخص

         إن عالمنا اليوم مدفوع بالإنجازات التكنولوجية والمعلوماتية التي تتميز أساساً بنشر الصور الرقمية وان أكثر أشكال نقل المعلومات شيوعاً هي المجلات والصحف والمجلات العلمية أو الإنترنت. رافق هذا الإستخدام الهائل لتقنية الصور تطوراً في أدوات التحرير لمعالجة الصور مما يجعل تعديل الصور وتحريرها في غاية البساطة، في أيامنا هذه أصبح تداول هذه الصور المزيفة التي تشوه الحقيقة شائعاً عن قصد أو عن غير قصد. هناك العديد من أنواع تزوير الصور الرقمية لكن النوع الأكثر أهمية وشعبية هو (نقل – نسخ) إنه سهل التنفيذ ويصعب اكتشافه. نظراً لأن المنطقة التي تم نسخها ولصقها من نفس الصورة ويتم تنفيذ عمليات ما بعد المعالجة على الصورة بأكملها فإن خصائص المنطقة المنسوخة متوافقة مع تلك الصورة لذلك فإن الطرق المعتادة للكشف عن عدم التوافق التي تستخدم القياسات الإحصائية لمقارنة عدة أجزاء من الصورة ليست مفيدة للكشف عن التزويرعلاوةً على ذلك يمكن إجراء عدة أنواع من التحويلات لجعل الصورة التي تم العبث بها أكثر واقعية.

1.  INTRODUCTION

          Since every information we get in our daily routine has been in a digital form with increasing technological advancement, it has brought a major issue of security. With the latest development of powerful algorithms and technology for manipulating digital images has made a very difficult task for determining a genuine image. The Digital image can be manipulated using many techniques and software like transformation, scaling, filtering, cropping, blurring, Photoshop, coral draw, and others. Image forgery has indeed become a major challenge for institutions as well as individuals. The materialization of digital images necessary for much implementation .

          Digital Images fraud detected by using two diverse approaches. These are effective approach and unassertive approach. The effective approach is two types, one is watermark and another is secret writing. Basic idea in watermark is to embed some information in digital images so that it cannot be misused or owned by others and message validation. Steganography is technique to hide a message within a digital image, to protect the privacy of the data.

Fig 1: Classification of different types of Image forgery

          The unassertive approach doesn’t demand any advance information in image as can examine image for any effect of manipulation during preprocessing step. Figure 1 [1] gives a complete classification of different types of image forgery.

2. Types of photo fraud

          Move forgery or reproduction, in which one area in image is being copy and paste into another area in same image So that it becomes an original image [2]. Since, the part in image being copy in same image has the same texture, noise, and color as rest of the image which makes it difficult to determine if it an original image or a forged image. This kind of image forgery is common nowadays, as it can use for the altercations on the image related to crime evidence, magazines etc. Many methods have been developed for the detection of this forgery which involves extracting features or dividing the image into blocks, which we will present in the next section.

          (Left) (Right)

Fig  2: (Left) Original Image (Right) Rigged Image

3. Image Magnification

   In Image magnification, several images merged both to create fake scene. Image magnification can easily be detected by looking at the boundaries of the image as it contains different image regions and they all have a different texture, color, and noise which can easily be detected. Several algorithms such as [3], [4] which effectively detect the image splicing forgery have been developed.

          (Left) (Right)

Fig 3 – (Left) Magnification image (Right) The original Image

4. Studying counterfeiting detection techniques

The following steps are common in the algorithms for the detection of the tampered region.

1. A forged color image of size P x Q is changed into grayscale or it is used as it is as per the algorithm.
2. Block-based methods. Some algorithms involve separating an image into blocks by sliding a small window over the image. Then applying the forgery detection algorithm over those blocks.
3. Key point based methods. There are several algorithms by which features are extracted which will be described later in this paper, which extracts the points which make the image unique. If an image is being forged then that image contain similar feature points. These features are stored as a feature vector or in a matrix depending on the algorithm.
4. Many morphological operations are apply this to detect the fraud area.

In[5] developed a new algorithm in order to account for low contrast segments in a forged image. It is a combination of DCT with PCA. First, PCA is applied to the image from which important features are extracted. Afterward, a window of size a xa slide over image for cut image coinciding lump. Local contrast for each block is calculated, those blocks which exceeded the fixed contrast are kept. 2D-DCT apply each lump and local quality matrix is extracted. This feature matrix is lexicographically sorted, and autocorrelation is evaluated. If the autocorrelation exceeds a threshold than those blocks are considered to be duplicates.[6] proposed an approach using discrete wavelet transform (DWT), in this approach consider a grayscale image or convert a color image into a grayscale image. Then, DWT transform apply the converted image and divide into sub-images, these sub-images are labeled as LL¹, LH¹, HL¹, and HH¹. A block size bxb is slide on these sub-images from LL to HH. DWT transform is calculated and stored in a lexicographical manner in the matrix. Normalized shift vector is calculated for the matched pairs and it is compared with the user-defined threshold to determine if the region is being copied.[7] uses a hybrid approach which involves DWT with DCT. In this approach consider a grayscale image or convert a color image into a grayscale image and apply DWT to divide the image into sub-images. Slide a window of size n x n over those sub-images resulting in K blocks. DCT is applied on the rows of the blocks to reduce the vector length and matrix formed H. The matrix lexico graphically sort Normalized shift vector is calculated for the matched pairs and it is compared with the user-defined threshold to determine if the region is being copied. In [8] approach Dyadic Wavelet Transform is used i.e. DyWT. Apply DyWT on the image to get LL² and HH² sub-bands. Analyze the pattern for each segment, and calculate the Euclidean between the pair of patterns. Check if the distance is less than the threshold T. If yes, then that region is marked as forged.[9] has proposed an approach which involves SURF and DyWT, which is identical to SURF and DWT [10] but DyWT is used as DWT is not shift-invariant. In this approach DyWT is performed into an image which divides the image into sub-images, key points and features are extracted by applying SURF. Feature descriptor vector of SURF is obtained and determine the similar feature descriptors. Finally, mark the forged regions.

[11] proposed a key point technique by using SIFT. SIFT involves four major step first, Scaled spaced extreme detection which finds the interest points using Laplacian of Gaussian (LOG). Second, key points select based on the measure stability. Third, Depend on scale, a neighborhood point chosen around key point position. Fourth, a 16×16 neighborhood is selected near the key point and it is separated into 16 sub-blocks of 4×4 size. Then, the Key point between similar image is matched.[12] proposed an algorithm involving domestic duo and neighborhood aggregation. In this algorithm a colored image separated R, B and G color components and these components are divided into blocks. Extract matrimonial blocks which are common in the three components, if duplicates were found then create sub-blocks using clustering and should a visual result.[13] proposed an algorithm for both key point regions and smooth regions. Initially, Simple Linear Iterative Clustering [14] is applied to the image separated image different block. Then SIFT used extract a key point from each block, the number of key points in a region is divided by the total number of pixels in that region to determine if it is a smooth region or a key point region. If it is detected as a key point region then the duplicate part is marked by using multiple key point matching [15] and RANSAC [16] is used to filter outliers. If it is a smooth region then Zernike moments [17] is used to detect the copied part.[18] presented an approach using SWT and SVD. Initially, the colored image is changed into grayscale and SWT is applied to separate the image into sub-bands LL1, LH1, HL1, and HH1. Then, the LL1 band is separated into coinciding blocks since LL1 gives the smoothest version of the image. SVD is used to extract a feature of the image and store it in a matrix, this feature matrix is lexicographically sorted. Check for similar features and calculate the Euclidian distance between then and store it in list L. The block pairs in list L are compared with threshold value T_s which is the minimum distance between duplicated regions. If it any block pair distance is more than T_s it is considered as a forged region.[19]  proposed an algorithm which uses 2D-DCT and SVD. First, the image is separated into coinciding blocks of equal size. 2D-DCT is applied to each block and DCT quantized coefficients are obtained. SVD is applied to extract features on each quantized block and stored in a lexicographically order in a matrix. Euclidian distance is calculated between matched pairs and compared with a threshold. After removing the disturbed matching output image is obtained containing marked forged region.[20] presented an algorithm using non-negative matrix factorization in which we normalize the original input image, then a block size BxB is slide on the normalized image divide image overlapping block. NMF apply to each block for feature extraction and it is lexicographically stored in a matrix. Then the hamming distance between two similar matrices is calculated and matched with the threshold T. If it is less than the threshold T, then it is considered as similar blocks and clustered together according to their displacements. Those blocks are marked as a duplicate region.[21] presented a robust and efficient technique in which it involves a block of size a x ais slide on image separate it coinciding block, and seven characteristic features are calculated for each block C_j (j=1,2,3…7) where C₁, C_2,and C₃red, green and blue components. C₄, C₅, C_6, and C₇ are the characteristics feature of Y channel which is a combination of R,G andB. These characteristic features stored vector V for each block separately and saved in an array A. The array A is lexicographically sorted and similarity is calculated between two vectors if it is greater than a threshold L it is recorded. A histogram is prepared with the recorded vectors, greater than threshold L and choose the main vector, d if any vector differs too much from d, it is discarded and remaining vectors are put in a binary image with the forged region set to white and rest of the region is set to black.

CONCLUSION

   Copy-move forgery has a huge impact on our day to day life. Steps that should be followed to detect the copy-move forgery. A detailed description of algorithms used to detect copy-move forgeries while some algorithms are not very effective to detect the forged region and others have the major problems which requires the computational time to be reduced, increase the accuracy, robustness against various geometric transformations. But some algorithm have achieved the mentioned goals up to certain extent like the algorithm which detects the forgery in both key point region and the smooth region is considered the best among the discussed methods. Also in future, there is a need to develop a robust and effective algorithm to detect Image forgery.

REFERENCES

• Jung, Ki-Hyun. (2016). Block-Based Authentication Scheme for Forgery Attacks on Digital Images. Advanced Science Letters. 22. 2583-2587. 10.1166/asl.2016.7784.
• Kanagavalli and L. Latha, “A survey of copy-move image forgery detection techniques,” 2017 International Conference on Inventive Systems and Control (ICISC), Coimbatore, 2017, pp. 1-6.
• A. Alahmadi, M. Hussain, H. Aboalsamh, G. Muhammad and G. Bebis, “Splicing image forgery detection based on DCT and Local Binary Pattern,” 2013 IEEE Global Conference on Signal and Information Processing, Austin, TX, 2013, pp. 253-256.
• Hakimi, M. Hariri, and F. GharehBaghi, “Image splicing forgery detection using local binary pattern and discrete wavelet transform,” 2015 2nd International Conference on Knowledge-Based Engineering and Innovation (KBEI), Tehran, 2015, pp. 1074-1077.
• Hilal, T. Hamzeh and S. Chantaf, “Copy-move forgery detection using principal component analysis and discrete cosine transform,” 2017 Sensors Networks Smart and Emerging Technologies (SENSET), Beirut, 2017, pp. 1-4.
• F. Hashmi, A. R. Hambarde and A. G. Keskar, “Copy move forgery detection using DWT and SIFT features,” 2013 13th International Conference on Intelligent Systems Design and Applications, Bangi, 2013, pp. 188193.
• Ghorbani, M. Firouzmand, and A. Faraahi, “DWT-DCT (QCD) based copy-move image forgery detection,” 2011 18th International Conference on Systems, Signals, and Image Processing, Sarajevo, 2011, pp. 1-4.
• Muhammad, M. Hussain, G. Muhammad and G. Bebis, “Copy-Move Forgery Detection Using Dyadic Wavelet Transform,” 2011 Eighth International Conference Computer Graphics, Imaging and Visualization, Singapore, 2011, pp. 103-108.
• F. Hashmi, V. Anandand A. G. Keskar, “A copy-move image forgery detection based on speeded up robust feature transform and Wavelet Transforms,” 2014 International Conference on Computer and Communication Technology (ICCCT), Allahabad, 2014, pp. 147-152.
• Alamro, Loai&Yusoff, Nooraini. (2017). Copy-move forgery detection using integrated DWT and SURF. Journal of Telecommunication, Electronic and Computer Engineering (JTEC). 9. 67-71.
• Prasad and B. Ramkumar, “Passive copy-move forgery detection using SIFT, HOG and SURF features,” 2016 IEEE International Conference on Recent Trends in Electronics, Information & Communication Technology (RTEICT), Bangalore, 2016, pp. 706-710.
• AlSawadi, G. Muhammad, M. Hussain and G. Bebis, “Copy-Move Image Forgery Detection Using Local Binary Pattern and Neighborhood Clustering,” 2013 European Modelling Symposium, Manchester, 2013, pp. 249-254.
• R. Resmi and S. Vishnukumar, “A novel segmentation based copy-move forgery detection in digital images,” 2017 International Conference on Networks & Advances in Computational Technologies (NetACT), Thiruvananthapuram, 2017, pp. 346-350.
• K. S. Varma, N. K. K. Rao, K. K. Raju, and G. P. S. Varma, “Pixel-Based Classification Using Support Vector Machine Classifier,” 2016 IEEE 6th International Conference on Advanced Computing (IACC), Bhimavaram, 2016, pp. 51-55.
• Debbarma, A. B. Singh and K. M. Singh, “Keypoints based copy-move forgery detection of digital images,” 2014 International Conference on Informatics, Electronics & Vision (ICIEV), Dhaka, 2014, pp. 1-5.
• M. Isaac and M. Wilscy, “A key point based copy-move forgery detection using HOG features,” 2016 International Conference on Circuit, Power and Computing 19
• Technologies (ICCPCT), Nagercoil, 2016, pp. 1-6.

RyuSJ., Lee MJ., Lee HK. (2010) Detection of Copy-Rotate-Move Forgery Using Zernike Moments. In:
Böhme R., Fong P.W.L., Safavi-Naini R. (eds) Information Hiding. IH 2010. Lecture Notes in Computer
Science, vol 6387. Springer, Berlin, Heidelberg.

• Dixit, R. Naskarand S. Mishra, “Blur-invariant copy-move forgery detection technique with improved detection accuracy utilizing SWT-SVD,” in IET Image Processing, vol. 11, no. 5, pp. 301-309, 5 2017.
• Moradi-Gharghani and M. Nasri, “A new block-based copy-move forgery detection method in digital images,” 2016 International Conference on Communication and Signal Processing (ICCSP), Melmaruvathur, 2016, pp. 1208-1212.
• Yao, T. Qiao, Z. Tang, Y. Zhao and H. Mao, “Detecting Copy-Move Forgery Using Non-negative Matrix Factorization,” 2011 Third International Conference on Multimedia Information Networking and Security, Shanghai, 2011, pp. 591-594.
• Weiqi Luo, Jiwu Huang and GuopingQiu, “Robust Detection of Region-Duplication Forgery in Digital Image,” 18th International Conference on Pattern Recognition (ICPR’06), Hong Kong, 2006, pp. 746-749.