Volume 5, Issue 5, May 25 ISSN: 2277 2X International Journal of Advanced Research in Computer Science and Software Engineering Research Paper Available online at: www.ijarcsse.com Image Authentication Using DWT Approach Tanu Dua, Dr. Bhupesh Kumar Singh CSE Department, M.D.U Rohtak, Haryana, India Abstract is a digital code irremovably robustly and imperceptibly embedded in the host data and typically contains information about origin status and destination of the data. In this paper we proposed an algorithm that embeds the watermark information without much distortion to images and thereby preserving the image quality. The watermark is embedded and extracted in the transform domain using Discrete Wavelet Transform (DWT) technique and it is checked under various noise attacks such as salt and pepper, and additional white. To do so various samples of coloured and grayscale images has taken and the performance is evaluated on the basis of similarity factor and peak to noise ratio as in []. The SNR and PSNR are used to measure the quality of an image after the reconstruction. Keywords, Discrete Wavelet Transform, Attacks, Decorrelation, I. INTRODUCTION Internet is an excellent distribution system for digital media because it is inexpensive, eliminates warehousing and stock, and delivery is almost instantaneous. However, content owners also see a high risk of piracy. This risk of piracy is exacerbated by the proliferation of high capacity digital recording devices. Thus content owners are eagerly seeking technologies that promise to protect their rights. The first technology content owners turn to is cryptography. Cryptography is probably the most common method of protecting digital content which can protect content in transit but once decrypted the content has no further protection. Thus there is a strong need for an alternative or complement to cryptography (a technology that can protect content even after it is decrypted). Image authentication techniques have recently gained great attention due to its importance for a large number of multimedia applications. Digital images are increasingly transmitted over non-secure channels such as the Internet. Therefore, military, medical and quality control images must be protected against attempts to manipulate them; such manipulations could tamper the decisions based on these images. To protect the authenticity of multimedia images, several approaches have been proposed [2].Digital watermarking is a technique which embeds additional information called digital signature or watermark into the digital content in order to secure it. A watermark is a hidden signal added to images that can be detected or extracted later to make some affirmation about the host image. ing include broadcast monitoring, transaction tracking, authentication, copy control, and device control. II. RELATED WORK DWT The wavelet transformation is a mathematical tool that can examine an image in time and frequency domains, simultaneously. DWT is simple and fast transformation approach that translates an image from spatial domain to frequency domain. The DWT provides a number of powerful image processing algorithms including noise reduction, edge detection, and compression. The transformed image is obtained by repeatedly filtering for the image on a row-byrow and column-by-column basis. Wavelet transform decomposes an image into a set of band limited components which can be reassembled to reconstruct the original image without error. Since the bandwidth of the resulting coefficient sets is smaller than that of the original image, the coefficient sets can be down sampled without loss of information. Reconstruction of the original signal is accomplished by up sampling, filtering and summing the individual sub bands. For two-dimensional (2-D images), applying DWT corresponds to processing the image by 2-D filters in each dimension. The filters divide the input image into four non-overlapping multi-resolution coefficient sets, a lower resolution approximation image (LL) as well as horizontal (HL), vertical (LH) and diagonal (HH) detail components. The subband LL represents the coarse-scale DWT coefficients while the coefficient sets LH, HL and HH represent the finescale of DWT coefficients[]. To obtain the next coarser scale of wavelet coefficients, the sub-band LL is further processed until some final scale N is reached. When N is reached we will have N+ coefficient sets consisting of the multi-resolution coefficient sets LLN and LHX, HLX and HHX where x ranges from until N. Due to its excellent spatio-frequency localization properties, the DWT is very suitable to identify the areas in the host image where a watermark can be embedded effectively. In particular, this property allows the exploitation of the masking effect of the human visual system such that if a DWT coefficient is modified only the region corresponding to that coefficient will be modified. In general most of the image energy is concentrated at the lower frequency coefficient sets LLx and therefore embedding watermarks in these coefficient sets may degrade the image significantly. Embedding in the low frequency coefficient sets, however, could increase robustness significantly [][5]. 25, IJARCSSE All Rights Reserved Page 9
May- 25, pp. 9-97 According to [], there is the DWT advantage over DCT as:. Input code is not divided into non-overlapping 2-D blocks, it has higher compression ratios avoid blocking artifacts. 2. Allows better localization both in time and spatial frequency domain.. Transformation of the whole image introduces inherent scaling.. Better identification of which data is relevant to human perception higher compression ratio. Figure: Methods of DWT 2. -D Discrete Wavelet Transform Wavelet can represent a signal in time-frequency domain. Analyzing a signal with this kind of representation gives more information about the when and where of different frequency components. In other words, wavelet transform not only transforms the time domain representation of a signal to the frequency domain representation, but also preserves spatial information in the transform. This feature enhances the image quality especially for the low bit rate representation. The DWT is a multi-resolution technique that can analyze different frequencies by different resolutions. The low-pass and high-pass filter pair is known as analysis filter-bank. An example of a low-pass filter is h (n) = (, 2,, 2, )/, which is symmetric and has five integer coefficients. An example of a high-pass filter is h (n) = (, 2, )/2, which is also symmetric and has three integer coefficients. These low- and high pass filters are used in the (5, ) filter transform. After applying the -D DWT on a signal that has been decomposed into two bands, the low-pass outputs are still highly correlated, and can be subjected to another stage of two-band decomposition to achieve additional decorrelation (In image processing decorrelation techniques can be used to enhance or stretch, colour differences found in each pixel of an image )[9]. In addition, the -D DWT can be easily extended to two dimensions (2-D) by applying the filter-bank in a separable manner. At each level of the wavelet decomposition, each row of a 2-D image is first transformed using a -D horizontal analysis filter-bank (h, h). The same filter-bank is then applied vertically to each column of the filtered and sub-sampled data. 2.2 2-D Discrete Wavelet Transform The 2-D DWT is computed by performing low-pass and high-pass filtering of the image pixels as shown in Figure 2. In this figure, the low-pass and high-pass filters are denoted by h and g, respectively. This figure depicts the three levels of the 2-D DWT decomposition. At each level, the high-pass filter generates detailed image pixels information, while the low-pass filter produces the coarse approximations of the input image. At the end of each low-pass and high-pass filtering, the outputs are down-sampled by two ( 2). In order to provide 2-D DWT, -D DWT is applied twice in both horizontal and vertical filtering. In other words, a 2-D DWT can be performed by first performing a-d DWT on each row, which is referred to as horizontal filtering, of the image followed by a -D DWT on each column, which is called vertical filtering [7]. 25, IJARCSSE All Rights Reserved Page 9
May- 25, pp. 9-97 Figure2: Three level 2-D DWT decomposition of an input image using filter. The below Figure illustrates the first decomposition level (d = ). In this level the original image is decomposed into four sub-bands that carry the frequency information in both the horizontal and vertical directions. In order to form multiple decomposition levels, the algorithm is applied recursively to the LL sub-band. The second (d = 2) and third (d = ) decomposition levels as well as the layout of the different bands are shown in Figure illustrates. There are different approaches to implement 2-D DWT such as traditional convolution-based and lifting scheme methods. The convolutional methods apply filtering by multiplying the filter coefficients with the input samples and accumulating the results. Their implementation is similar to the Finite Impulse Response (FIR) implementation. This kind of implementation needs a large number of computations. Figure: Different sub-bands after first decomposition level. Figure :Sub-bands after second and third decomposition levels. 2. Lifting Scheme The lifting scheme has been proposed for the efficient implementation of the 2-D DWT. The lifting approaches need 5% less computations than the FIR approaches. The basic idea of the lifting scheme is to use the correlation in the image pixels values to remove the redundancy. The lifting scheme has three phases, namely, split, predict, and update, as illustrated in Figure 5. In the split stage, the input sequence is split into two sub-sequences Figure5: Three different phases in the lifting scheme. consisting of the even and odd samples. In the predict and update stages, the high-pass and low-pass values are computed, respectively. Lifting scheme has advantages than traditional convolution-based algorithms. Few of them are following: 25, IJARCSSE All Rights Reserved Page 9
Dua et al., International Journal of Advanced Research in Computer Science and Software Engineering 5(5), May- 25, pp. 9-97 Lifting scheme algorithms provide more performance improvement than traditional convolution-based implementation. The number of operation in the lifting scheme approaches is almost one half of that in the convolution-based approaches. Such reduction in the computational complexity makes lifting scheme filters attractive for high-performance implementation. In lifting scheme implementation, there is no need to manage the borders of an image, while in the traditional convolution-based approaches it is necessary. Lifting scheme approaches can be implemented in-place. This means that the DWT can be processed without using auxiliary array as is necessary in the traditional convolution-based implementation. All operations within one lifting step can almost be performed in parallel. With lifting transforms it is easy to build non linear wavelet transforms. For instance, wavelet transforms that map integers to integers. Such transformations are more important for hardware implementation and for lossless image compression. The inverse transform of lifting scheme is straight forward, it easily inverts the order of functionality. So the same operations and resources could be reused to implement IDWT. One example of this group is the integer-to-integer (5, ) lifting scheme ((5, ) lifting. 2. Convolutional Methods The convolutional methods apply filtering by multiplying the filter coefficients with the input samples and accumulating the results. The Daubechies transform with four coefficients (Daub-) and the Cohen, Daubechies and Feauveau 9/7 filter (CDF-9/7) are examples of this category. 2.5 2 2 Haar Transform The 2 2 Haar transform is sometimes referred to as a wavelet. A 2-D Haar transform can be performed by first performing a -D Haar transform on each row (horizontal Haar transform) followed by a -D Haar transform on each column (vertical Haar transform). This transform is used to decompose an image into four different bands. The -D 2 2 Haar transform replaces adjacent pixel values with their sums and differences. An example of the 2-D 2 2 Haar transform for a image is depicted in Figure. This transform generates four different subbands of low-pass and high-pass value [7]. There are different algorithms to traverse an image to implement these transforms, namely Row-Column Wavelet Transform (RCWT) and Line-Based Wavelet Transform (LBWT). 2. Row-Column Wavelet Transform In the RCWT approach, the 2-D DWT is divided into two -D DWTs, namely horizontal and vertical filtering. Horizontal filtering processes the rows of the original image and stores the wavelet coefficients in an auxiliary matrix. Thereafter, the vertical filtering phase processes the columns of the auxiliary matrix and stores the results back in the original matrix. In other words, this algorithm requires that all lines are horizontally filtered before the vertical filtering starts. The computational complexity of both horizontal and vertical filtering is the same. As shown in Figure both horizontal and vertical filtering. Each of these filtering is applied separately. Each of these N M matrices requires NMc bytes of memory, where c denotes the number of bytes required to represent one wavelet coefficient. Figure: 2-D 2 2 Haar transform using two -D horizontal and vertical Haar transform. 25, IJARCSSE All Rights Reserved Page 92
May- 25, pp. 9-97 2.7 Line-Based Wavelet Transform In the line-based wavelet transform approach, the vertical filtering starts as soon as a sufficient number of lines, as determined by the filter length, have been horizontally filtered. In other words, the LBWT algorithm uses a single loop to process both rows and columns together. This technique computes the 2-D DWT of an N M image by the following stages. First, the horizontal filtering filters L rows, where L is the filter length, and stores the low-pass and high-pass values interleaved in an L M buffer. Thereafter, the columns of this small are filtered. This produces two wavelet coefficients rows, which are stored in different subbands in an auxiliary matrix in the order expected by the quantization step. Finally, these stages are repeated to process all rows and columns. The LBWT algorithm is illustrated in Figure7. Figure 7: The line-based wavelet transform approach processes both rows and columns in a single loop. Factors affecting digital watermark strength When we embedding a watermark along with the intensity, the strength of a digital watermark is also affected by the following factors: Image variations/randomness: The successful embedding of a digital watermark is dependent on the variation and randomness present in the pixels making up the image. For example, if you are working with an image that contains more flat color regions than detailed areas, you may want to choose a higher digital watermark strength so that the watermark will overcome the limitations of the specific image. This may result in a more visible digital watermark, but in some situations that is an acceptable trade-off, as mentioned above. Image size: As far as possible the s should be immune to Image size. Compression: Saving the watermarked image in a compressed format may affect the durability of the digital watermark. The following factors will influence the impact that lossy compression has on digital watermark survival: Level of image compression: Lossy compression degrades the image to some extent, depending upon the quality setting chosen when saving in compressed format; most digital watermarks will survive as long as a moderate level of compression is used. Visibility/durability setting used when embedding a digital watermark: The higher the intensity setting, the better the chances the digital watermark will survive compression. Again, a higher-intensity digital watermark provides more data-to-survive compression. Since the visual quality of compressed images is often somewhat compromised anyway, generally a higher watermark intensity setting yields quite acceptable results. Image size: The greater the number of pixels in the image, the more the digital watermark can be repeated throughout it; the recommended minimum size for an image that will be compressed is 25 x 25 pixels. The larger the image, the better the digital watermark will survive compression. Randomness of image data: The more randomness and/or color variation in an image, the better; a flat color space with little gradation may not survive well, while an image with more detail and contrast will fare better. Since a digital watermark is applied more strongly within areas of high contrast or variation, an image that contains more contrast and/or variation than others will contain more digital watermark data and thus stand a better chance of surviving compression. The following two quality measures have been considered to evaluate the performance of digital watermarking techniques Similarity Factor (SMF): After successful recovery of watermark from the watermarked content, the recovered watermark has to compare with original watermark for the evaluation of the whole process. Similarity Factor defined as the co-relation between the original watermark ( W o ) and recovered watermark (W r ) using the following equation number (). The Value of the Similarity Factor (SMF) will be between and. SMF is defined as: () SMF ( N i Wo Wr) ( N i Wo^2 N i Wr^2) 25, IJARCSSE All Rights Reserved Page 9
May- 25, pp. 9-97 where, N = Total No. of Pixels in the watermark image. Peak Signal to Ratio (PSNR): Peak Signal to Ratio is an expression for the ratio between the maximum possible value (power) of a signal and the power of distorting noise that affects the quality of its representation. PSNR is usually expressed in terms of the logarithmic decibel scale. The PSNR is most commonly used as a measure of quality of reconstruction of lossy compression codec. The signal in this case is the original data, and the noise is the error introduced by compression. The general accepted values are values more than db. It is most easily defined via the mean squared error (MSE) with following equation number (2) for two m * n images I and K where one of the images is considered a noisy approximation (here image k) of the other is defined in following equation no. () as in []s. (2) () MSE (/ n m) m n i j PSNR 2 log(im MSE ) [ I( i, j) K( i, j)]^2 III. PROPOSED METHOD DWT is used to transform the image into sub band coefficients and logistic chaotic sequence is utilized to scramble the watermark before it is embedded. The watermark is embedded into the important sub band coefficients taking into account the characteristics of Human Visual System (HVS)[]. embedding procedure by DWT using 2x2 blocks As we have grayscale image for embedding the watermark, we can straight away decompose the original image (without splitting it in Y,Cb,Cr bands). Next, out of the decomposed image, we divide the HL band into smaller blocks of 2x2. Then, we process the HL band, block by block. Mean for each block is calculated and is modified according to the watermarking bits. The generated watermarked HL plane coefficients are constructed, which are later used to construct the watermarked image. Inverse DWT is applied to modified coefficients and the remaining unaffected bands. 25, IJARCSSE All Rights Reserved Page 9
May- 25, pp. 9-97 extraction procedure for DWT using 2x2 blocks For the extraction of watermark, the exact opposite of the embedding process is done. The DWT transformation of the image is carried out, then, averages for each of the smaller 2x2 blocks of the resulting HL band are found. Average of absolute values and reconstruction of the watermark follows. If the average is odd, the corresponding bit is set, else, it is cleared. This way, we obtain the randomized watermark image. Finally, we use secret keys to get the actual watermarking image using the extracted randomized watermark. IV. EXPERIMENTAL RESULTS In order to test the performance of the proposed system.we have performed the experiment with different samples,five coloured and five gray scale and considered four different noise-attacks. We used [x] size standard coloured and gray scale images as a samples using MATLAB as a tool. The observation table is made for DCT and DWT. The values of DCT is discussed in []. Samples Nature s Flower Butterfly Palm Ship Sno Attacks DCT DWT DCT DWT DCT DWT DCT DWT DCT DWT No-.9977.999.9957.997.99.995.7..9992.999 Salt&Pepper.92.975.9595.95.9595.975..5.9.9 2 Speckle.7.725.292..292..2..77.75 For Add..2.75.5.25.9.52.752...2 White For..75.2.97.2.5.7.7.9.57 Figure.: Observation Tables of Different Coloured Images with SMF Samples Eye Girl Tree Clock Bird Sno Attacks DCT DWT DCT DWT DCT DWT DCT DWT DCT DWT No-.9979.99.99.99.99.997.59.7.99.99 Salt&Pepper.95.92.9575.97.9572.972.27.7.92.977 2 Speckle.79.79.95.925.579.79.5.9.7.799 For Add..92..5.97..25..92.99.2 White For..75.2.99..5.2.9..5 Figure.2: Observation Tables of Different Gray scale Samples with SMF Samples Nature s Flower Butterfly Palm Ship Sn Attacks DCT DWT DCT DWT DCT DWT DCT DWT DCT DWT o No-.29 7 2.9.25 7.5 9..9.7 2.7 2 7.95 9.5 2 Salt&Pepper.79.999. 5.7 9.9. 9.595. 5.92 7.2 2 Speckle. 2..97 5.2 9.2. 9..7.77 5.2 7 ForAdd.Whi te 27.5 2. 27.7 2.952 27.9 27.99 27.75 27.92 9 27.55 2 29.2 For 9.7..2.999.27.95.9. 2 Figure.: Observation Tables of Different Coloured Images with PSNR.75 2.9 Samples Eye Girl Tree Clock Bird Sn o Attacks DCT DWT DCT DWT DCT DWT DCT DWT DCT DWT 25, IJARCSSE All Rights Reserved Page 95
May- 25, pp. 9-97 No- 7.95.7.5 5.7.7.9..999 7.5.9 7 7 5 Salt&Pepp 5.92.2.7.79.5.7 5.95 7. 7.22 9. er 2 5 2 Speckle.77.277. 5.2 9.7.2 5.7.97 7.222.2 2 7 7 For Add. 27.55 2.72 2.9 2.9 27. 2. 2.2 29.9 27.7 2.25 White 2 9 2 7 For.75 2.57.5.5 9.7 7 9.52 7 5.52 5.7 Figure.: Observation Tables of Different Gray scale Images with PSNR.2 7.92 Original Image Original Randomize ed Image Extracted Randomize Extracted Original Figure.5: Conformance Experiment of the ed Images Original Image Original Randomize ed Image Extracted Randomize Extracted Original 25, IJARCSSE All Rights Reserved Page 9
May- 25, pp. 9-97 Figure.: Conformance Experiment of the ed Images V. CONCLUSION Digital watermarks are efficient in providing ownership identification for images. must be robust and imperceptible. Robustness of watermark can be explained in terms of successful recovery of watermark from recovered content which may contain different types of noises and compression effects. After recovering the watermark, the recovered and original watermarks are compared by calculating of similarity factor of these two watermarks. If the similarity factor is closer to one than we can conclude that the content is original and/or authenticated.. Even in the absence of noise, the extracted watermark doesn t come out to be exactly same as original one, i.e. similarity is never equal to, though very close to it. 2. Introduction of noise challenges the watermarking procedure to a greater extent than any other noise/attack.. Increasing the variance of distribution further degrades the quality of the watermark extracted.. Similarity factor between original and extracted watermark decreases as we go on increasing the value of the variance in the case of the and Speckle.. In most cases, the similarity factor stays over.7 and the PSN ratio well above 25dB, which are the criteria to judge the robustness of the watermarking algorithm. REFERENCES [] Dr. Bhupesh Kumar Singh, Tanu Dua : Image Authentication Using Digital ing,issn (e): 225 5 Volume, 5 Issue, April 25, International Journal of Computational Engineering Research (IJCER) [2] A. Haouzia: R. Noumeir (*) Electrical Engineering Department, École de Technologie Supérieure, Notre- Dame West, Montreal, Quebec, Canada, HC K [] Ekta Hans, Sheelu Sharma,Nitu Rani : A Review of Digital ing Techniques for Images Volume-, Issue-, June-2, ISSN No.: 225-75 International Journal of Engineering and Management Research [] Al-Haj: Combined DWT-DCT digital image watermarking, Journal of Computer Science (9): pp. 7-7, 27 [5] M. Chandra, S. Pandey: A DWT Domain Visible ing Techniques for Digital Images," International Conference on Electronics and Information Engineering, pp. V2-2 -V2-27, 2 [] D. P. R. K. V.Srinivasarao, G.V.H.Prasad, M.Prema Kumar, S.Ravichand, Discrete Cosine Transform Vs Discrete Wavelet Transform: An Objective Comparison of Image Compression Techniques for JPEG Encoder, International Journal of Advanced Engineering & Applications, Jan 2. [7] Hingoliwala H.A., An image compression by using haar wavelet transform, Advances in Computer Vision and Information Technology, 2. [] Yin Ke ; Wang Zhi-quan ; Liu Chang ; Yin Ke-xin, Digital ing Scheme Based on Chaotic Encryption and DWT, Information Engineering and Computer Science, 29. ICIECS 29. International Conference. [9] Asymptotic Decorrelation of Between-Scale Wavelet Coefficients, Peter F. Craigmile and Donald B. Percival, Department of Statistics, The Ohio State University, Cockins Hall, 95 Neil Avenue, Columbus. OH 2. Applied Physics Laboratory, Box 55, University of Washington, Seattle. WA 995 5. 25, IJARCSSE All Rights Reserved Page 97