Detection of Face Morphing Attacks based on Half-Toning Feature Extraction
1E.Gangadurai , 2Ummaneni Hemanth, 2D Vigneshwaran ,2R Sureshkumar ,2P Prakash
1Assistant Professor I 2UG Student, Department of Electronics and Communication Engineering Velammal Engineering College, Chennai - 600 066.(INDIA)
Abstract:
In All Applications, the need for Image Security is alarmingly increasing. Multiple areas use this Image Security system. Also, there is more research going in Image Manipulation and Image Processing. Here, we introduce a Halftoning method with LU Decomposition and self-Recovery in the Medical industry. To Find the tampering ratio, we need to check for Peak Signal to Noise Ratio (PSNR) value or Percentage. In this paper, we will be seeing the Correction of the image and recovery of the image. At first the input image is divided into 4*4 matrixes. Then LU is used to find out the changes in the original image. The recovery watermark is created in this halftoning method. For the analyze of standard of the output images, peak signal to noise ratio is utilized. The advantage in our method is contrasted with other variety of techniques, in a way that standard of the watermark images and recuperation rate of others were analyzed by exploratory outputs.
Keywords: Half Toning, LU Decomposition, Arnold Transformation, superimposing, morph detection.
I. INTRODUCTION
The evolution of electronic mobile appliances, like mobile phones and camera the images are obtained more and more easily. The potentiality of the image rewriting software become bigger. The originality and probity are so major in digital certainty that more analysts put attention on this subject. The watermarking methodology, as one of the attestation methods, can find the potentiality and rectify the modified part successfully, and bring the original input tamper free image. The functionality of image self-restoration has three main parts: the attestation particulars, the restored information to the image.
The information has to find out the originality of the input image effectively, and it can isolate the modified part exactly. The mapping performance can infix the data by changing the pixels, and the standard of the image. Thus an improved performance could be the key of planned algorithmic directive. The aim of the changed rule is to enhance the fixed image standard and the modified space restoration correctly. The methods of image self-restoration supported watermark is identified as two varieties from infixing field: the special rework embedding. This techniques can change the pixel values within the special domain. They are easy and effective. For instance the twin watermark to demonstrate the tampered image and acquire a decent recovery performance where as the massive tamper quantitative relation seems. The image is split into 2*2 pixels of blocks, and so the typical values of every box area unit accustomed build the restored data, that is infixed into the 2 LSB’s. The restoration performance in an exceedingly high tamper quantitative relation. To reduce the choice of building associate in nursing incorrect prediction, the tactic produces confirmation and bits from pixels where bits are repositioned. The confirmation bits area unit made from pixels those bits are regrouped. The playing code is employed to rebuild the authentication data . To boost the safety of these algorithms, Arnold rework is implied within the connection of blocks.
To increase the restoration presentation and take advantage of abstraction infixing, a replacement LU rotten halftoning theme for image authentication and self-restoration for medical usages. The projected theme locates image change at state additionally as recovers
the primary image. Variety image is split into 4×4 blocks and LU is implemented to figure out the change inside the first image. After that generates, the authentication watermarks, that supported XOR on non- overlapping blocks, later by using a halftoning method the restored watermark is created.
II. RELATED WORKS
In this section, first, many tamper identification and restoration methods are evaluated, which produced in recent years as relevant tasks. Then, the benefactions of the suggested techniques are introduced. Lee and Lin planned an successful double watermark for image tamper identification and restoration. Dual watermarking methods can increase the standard of the restored image. During this system, two prints of the watermark are implanted into two various places within the full image. Thus, it can give a second possibility to change modified block just in case one copy is terminated, but this system isn't ready to notice any subjects interfering that changes bits in 5 most important bits. Final outputs embellished flexibility for casing, eliminating, snipping, and changing tampering and endangered in case of collage, vector quantitation, and copy-move interfering. In another work, a probability-supported tampering identification method for digital images is produced by Hsu and Tu. This theme points to use applied mathematics to enhance the tamper finding rate. within the tamper identification phase, first, the watermarked image is clarified through the implanted within the image. The authors proposed an efficient self-embedding fragile watermarking for image tamper localization and recovery supported DCT. This scheme performed an improved tamper localization and recovery algorithm compared to previous methods. within the proposed scheme for enhancing the safety of the algorithm, a non-linear chaotic sequence is been used within the embedding phase, the watermark is generated by encoding DCT coefficients of every 2×2 block and conceal in another block consistent with the block mapping. An efficient singular value decomposition (SVD)–supported image tampering detection and self- restoration are produced by Dadkhah et al. to enhance the tamper detection ratio, a mixed block partitioning approach for 4×4 and 2×2 blocks is employed. The experimental outputs shows that the proposed theme is patronizing in terms of security, tamper focalization, and restoration rate, over the opposite fragile tamper identification and restoration themes during this process, three copies of the restoring watermark are implanted into various quadrants, which gives two possibility for recovery just in case one is destroyed.
III PROPOSED METHODOLOGY
A replacement LU decomposed halftoning scheme for image authentication and self-recovery for medical applications is described in this section. The proposed scheme both detects image tampering and restores the original image. The transformation inside the original image is worked out using LU after a number image is split into 4x4 blocks. The authentication watermarks are then produced, which are assisted by XOR operations on non- overlapping blocks, and the recovery watermark is then generated using a Halftoning technique.
Figure 1 Proposed architecture
The proposed scheme both detects image tampering and restores the original image. To figure out the transformation inside the original image, a number image is split into 4x4 blocks and LU decomposition is applied by inserting the traces of block-wise LU decomposition into the least significant bit (LSB) of the image pixels. To withstand the vector quantization attack, two authentication bits are used: block authentication and self- recovery bits. Arnold's transformation, which restores the first image even after a high tampering rate, is used to evaluate the insertion of self-recovery bits. Watermarking information based on LU decomposition increases image authentication and identifies the different areas of the watermarked image that have been targeted. Halftone is a reprographic technique that creates a gradient-like effect by using dots that differ in size or spacing to imitate continuous-tone imagery. The word "halftone" can also be used to refer to the image formed by this process. The halftone method reduces visual reproductions to a picture printed with only one colour of ink, in dots of varying size (pulse-width modulation) or spacing (frequency modulation), or both, where continuous- tone imagery includes an infinite number of colours or greys. The human eye interprets the patterned areas as smooth tones when the halftone dots are tiny, so this reproduction relies on an optical illusion. The established black-and-white film has only two colours on a microscopic level, rather than an infinite spectrum of continuous tones. See film grain for more information. Colour printing is made possible by repeating the halftone method for any subtractive colour – most often using the "CMYK colour model" – much as colour photography developed with the addition of filters and film layers. Ink's semi-opaque property allows halftone dots of different colours to create a full-colour optical effect.
Spatial and frequency filtering
The removal of halftone patterns and reconstruction of tone changes are the most time-consuming parts of the process. In the end, it will be necessary to recover details in order to improve image quality. There are a variety of halftoning algorithms, but the most common are ordered dithering, error diffusion, and optimization-based methods. It's critical to choose the right descreening strategy because different descreening strategies produce different patterns, and most inverse halftoning algorithms are designed for a particular pattern type. Because many algorithms are iterative and thus slow, time is another consideration. The application of a low-pass filter, either in the spatial or frequency domain, is the most straightforward way to eliminate halftone patterns. A Gaussian filter is a simple example. It discards high- frequency information, blurring the image and reducing the halftone pattern at the same time. When watching a halftone image, this is often similar to the blurring effect of our eyes. In any case, picking the right bandwidth is critical. A low bandwidth blurs edges, while a high
bandwidth creates a noisy image by not entirely removing the pattern. It is unable to rebuild fair edge information as a result of this trade-off..
Optimization-based filtering
The use of machine learning algorithms to support artificial neural networks is another option for inverse halftoning. These learning-based methods may identify the descreening method that comes the closest to being correct. The idea is to employ various strategies depending on the halftone picture. The approach should be varied even for different content within the same photograph. Convolutional neural networks are well-suited for tasks like object detection, allowing for descreening based on categories. They'll also use edge detection to strengthen the small print around the edges. However, these methods are constrained by the quality and completeness of the training data used. Unseen halftoning patterns that were not reflected in the training data are difficult to remove. Furthermore, the training process can be lengthy.
In contrast to other iterative methods, computing the inverse halftoning image is quick because it only takes one computational step.
LU decomposition
Assume A is a matrix. An LU factorization is the factorization of A into two factors – a lower triangular matrix L and an upper triangular matrix U – with proper row and/or column orderings or permutations:
A=LU (1) All elements above the diagonal are zero in the lower triangular matrix, and all elements below the diagonal are zero in the upper triangular matrix. For example, the LU decomposition of a 3 3 matrix A looks like this:
The factorization may fail to materialise if the matrix's ordering or permutations are incorrect. If A is nonsingular, this is often impossible (invertible). This is often a procedural issue. It's usually removed by rearranging the rows of A so that the permuted matrix's main element is nonzero. In subsequent factorization steps, an equivalent problem is often removed in an equivalent manner;
see the essential procedure below.
LU factorization with partial pivoting
For LU factorization, it appears that a correct permutation in rows (or columns) is sufficient. LU factorization with partial pivoting (LUP) is a term that is frequently used to refer to LU factorization with only row permutations.
PA=LU (2) where L and U are lower and upper triangular matrices, respectively, and P is a permutation matrix that reorders A's rows when left-multiplied. Each square matrices appears to be factorised frequently in this form, and thus the factorization is numerically stable in practise. As a result, LUP decomposition is a practical approach.
LU factorization with full pivoting
Both row and column permutations are involved in a LU factorization with full pivoting.
AQ=LU (3) where L, U, and P are the same as before, and Q is a permutation matrix that rearranges the columns of A.
LDU decomposition
An LDU decomposition may be a decomposition of the shape
A=LDU (4) where D is a square matrix and L and U are unit triangular matrices, which means that each of their diagonal entries is one. Although we assumed A to be a matrix earlier, these decompositions can all be applied to rectangular matrices as well. In this case, L and D are both square matrices with the same number of rows as A, and U has the same dimensions as A.
Upper triangular should be considered as having zero entries below the most diagonal, which begins in the upper left corner.
IV. RESULT AND DISCUSSION
In this section, the simulation results are simulated and implemented using MATLAB and the Xilinx software which is given in the following. The imaging results are taken in the MATLAB and the area comparison are taken from the Xilinx with the accurate synthesis report of existing and the proposed system.
4.1 OUTPUTS
Figure 2 Actually Send Input Image
Above figure 2 shows the input image for our embedding process.in this stage image converted into a grayscale image and resized to the required stage
Figure 3 Halftoned image
Above figure 3 shows the halftoned image of the input image by performing halftoning.
Figure 4 Scrambled and LSB numbered Image
Above Figure 4 shows scrambled images and LSB renumbered images for our embedding process.
In this stage image, authentication bits are identified using Arnold sampling
Figure 5 Input and Tamper Free Image
Above figure 5 shows the final tamper-free image in our embedding process.
Figure 6 Received Tamper Image
Above figure 6 shows the received image after an unknown attacker attacked.
Figure 7 Tamper detected Image
Above figure 7 shows tamper-identified images using our authentication bits.
(a)
(b)
Figure 8 (a) Tamper localization image and (b) Tamper localization percentage image Above figure 8 (a) and (b) shows the tamper localization and its percentage images using LU
decomposition.
Figure 9 Recovered Image
Above figure 9 shows the recovered image by performing LSB zero and LU decomposition 4.4 PERFORMANCE EVALUATION
The below table shows improved SNR calculation before and after embedding and extraction.
Table1PSNRvaluesofimages S.No Imageset PSNR
1 1 50.2302 2 2 51.0662 3 3 51.1394 4 4 50.7783
Figure 10 Comparison of Half Toning method with PRNU method
The values to plot the graph for PRNU method is obtained from the base paper that is from the Existing system. These values are the PSNR values that is compared with that of the
Proposed system. Also, When you see the plot, the Graph peak values keeps on changing and doesn’t have proper range of values. Whereas in Half Toning, we can say that it has a
proper range of PSNR from 50dB - 56dB ( So far in our test ).
V. CONCLUSION
Morphing of images is the most important technique that helps in copyright infringement of digital images. It is popularly said that the main problems in copyright infringement in this work let us consider the fall of regular superimposing techniques during the handling of alphanumeric attributes. Our project gives a LU and halftoning-based morph identification technique by grouped block method to provide high security and offer a supporting way to track the influenced areas in different images. Two verification bits as block verification and auto-recovery bits are used to overcome the vector quantization attack. The usage of verification makes it possible to retrieve the morphed region from the adjacent blocks, which finally raises the NCC and PSNR of the retrieved host. This method is more reliable than the previous methods, because the Error Detection Ratio of Half- Toning method is 98% whereas the Error Detection Ratio of PRNU method is 92 - 94%. So, Half-Toning method is the best way to detect the morph in images.
ACKNOWLEDGEMENT
First of all, we would thank God for guiding and helping us in every stage of our life. It is our pleasant duty to express deep sense of gratitude to our beloved Chairman Thiru M.V.Muthuramalingam for his constant guidance and support. We would like to express our heartfelt thanks to our respected Chief Executive Officer, Thiru. M.V.M.Velmurugan for his kind encouragement. We express our deepest gratitude and thanks to our beloved Principal, Dr. N.Duraipandian for his inspirational support during the course of the project. We would specially like to thank our Head of the Department of Electronics and Communication Engineering, Dr.S.Mary Joans, for her positive feedback. We are indebted to the undying support of our Supervisor, Assistant Professor II and E.GANGADURAI, Department of Electronics and Communication Engineering, Velammal Engineering College for her noble endeavour in encouraging our spirits to complete this project to the utmost satisfaction of our superior. Our special thanks to our Project Coordinator, Dr.K.Thilagam and Dr.C.Murukesh for her continuous support and ideas through the course of the project. Finally, we thank the teaching and non- teaching staff of our Department, parents and friends who helped us in the successful completion of this project.
REFERENCES
[1] Lee T-Y, Lin SD. Dual watermark for image tamper detection and recovery. Pattern Recognition, 2008: 41(11): 3497-3506. https://doi.org/10.1016/j.patcog.2008.05.003.
[2] Hsu C-S, Tu S-F. Probability based tampering detection scheme for digital images.
Optics Communication 2010; 283(9):1737-1743.
https://doi.org/10.1016/j.optcom.2009.12.073.
[3] Qian Z, Feng G, Zhang X, Wang S. Image self- embedding with high-quality restoration capability. Digital Signal Process. 2011;21(2):278–286.
https://doi.org/10.1016/j.dsp.2010.04.006
[4] Zhang J, Zhang Q, Lv H. a completely unique image tamper localization and recovery algorithm supported marketing technology. Optik-Internation Journal for Loght and Electronics 2013;124(23):6367–6371. https://doi.org/10.1016/j.ijleo.2013.05. 040.
[5] Dadkhah S, Manaf AA, Hori Y, Hassanien AE, Sadeghi S. an efficient and- based image tampering detection and self-recovery using active watermarking. Signal Process Image Commun. 2014;29(10):1197–1210. https://doi.org/10.1016/j.image.2014.
09.001.
[6] Zhang X, Xiao Y, Zhao Z. Self-embedding fragile watermarking supported DCT and fast fractal coding. Multimed Tools Appl. 2015;74(15):5767–5786.
[7] Shao-Hui Liu, Hong-Xun Yao, Wen Gao, Yong- Liang Liu, “An image fragile watermark scheme supported chaotic image pattern and pixel-pairs”, applied math and Computation, 185(2):869–882, 2007.
[8] Ninghui Li, Wenliang Du, and Dan Boneh, “Oblivious signature-based envelope”, Distributed Computing, 17(4):293–302, 2005.
[9] Toshihiko Matsuo and Kaoru Kurosawa, “On parallel hash functions supported block-ciphers”, IEICE Transactions on Fundamentals of Electronics, Communications, and Computer Sciences, 87(1):67–74,2004.
[10] Shah Suthaharan, “Fragile image watermarking employing a gradient image for improved localization and security”, Pattern Recognition Letters, 25(16):1893–1903, 2004.
[11] Chun-Shien Lu and H-Ym Liao, “Structural digital signature for image authentication: an incidental distortion resistant scheme”, IEEE Transactions on Multimedia, 5(2):161–173, 2003.
[12] Ping Wah Wong and Nasir Menon, “Secret and public key image watermarking schemes for image authentication and ownership verification”, IEEE Transactions on Image Processing, 10(10):1593–1601,2001.
[13] Matthew Holliman and Nasir Menon, “Counterfeiting attacks on oblivious block-wise independent invisible watermarking schemes”, IEEE Transactions on Image Processing, 9(3):432–441, 2000.
[14] Rahim, R., Murugan, S., Manikandan, R., & Kumar, A. (2021). Efficient Contourlet Transformation Technique for Despeckling of Polarimetric Synthetic Aperture Radar Image. Journal of Computational and Theoretical Nanoscience, 18(4), 1312-1320.
[15] Rahim, R., Murugan, S., Mostafa, R. R., Dubey, A. K., Regin, R., Kulkarni, V., &
Dhanalakshmi, K. S. (2020). Detecting the Phishing Attack Using Collaborative Approach and Secure Login through Dynamic Virtual Passwords. Webology, 17(2).
