A oint DT-DCT Based atermarking Technique for Avoiding Unauthorized Replication Kaushik Deb 1, Md. Sajib Al-Seraj 1, Mir Md. Saki Kowsar 1 and Iqbal Hasan Sarkar 1 1 Department of Computing Science and Engineering, Chittagong University of Engineering & Technology, Chittagong-4349, Bangladesh E-mail:debkaushik99@gmail.com Abstract Use of image has been increasing in many applications. ith the rapid development of the network multimedia systems, one can easily duplicate digital image. A digital image watermarking is a valid solution of protecting illegal manipulation of digital image. It embeds some secret information into the host image and this information uses for authentication. In this paper a joint DT and DCT based watermarking technique with low frequency watermarking with weighted correction is proposed. DT has excellent spatial localization, frequency spread and multi-resolution characteristics, which are similar to the theoretical models of the human visual system (HVS). DCT based watermarking techniques offer compression while DT based watermarking techniques offer scalability. These desirable properties are used in this combined watermarking technique. In the proposed method watermark bits are embedded in the low frequency band of each DCT block of selected DT sub-band. The weighted correction is also used to improve the imperceptibility. The extracting procedure reverses the embedding operations without the reference of the original image. Compared with the similar approach by DCT based approach and DT based approach, the experimental results show that the proposed algorithm apparently preserves superior image quality and robustness under various attacks such as PEG compression, cropping, sharping, contrast adjustments and so on. Keywords-Digital image watermarking; Discrete avelet Transform (DT); Discrete Cosine Transform (DCT). I. ITRODUCTIO In the recent years, it becomes a daily need to create copy, transmit and distribute digital data as a part of widespread multimedia technology by means of the orld ide eb. Hence copyright protection has become essential to avoid unauthorized replication problem. Digital image watermarking provides the essential mechanism for the ownership authentication. Image watermarking is the process of inserting hidden information in an image by introducing modifications of minimum perceptual disturbance. Robustness, perceptual transparency, capacity and blind watermarking are four essential factors to determine quality of watermarking scheme [1]. Image watermarking techniques proposed so far can be divided into two group s accordingly processing domain of host image. One is to modify the intensity value of the luminance in the spatial domain [-3] and the other is to change the image coefficient in a frequency domain [4-8]. Commonly used frequency-domain transforms include the Discrete avelet Transform (DT) and the Discrete Cosine Transform (DCT). However, DT has been used in digital image watermarking more frequently due to its excellent spatial localization and multi-resolution characteristics, which are similar to the theoretical models of the human visual system and DT gives perfect reconstruction of decomposed image. The DCT has special property that most of the visually significant information of the image is concentrated in just a few coefficient of the DCT. Moreover DCT based watermarking techniques offer compression while DT based watermarking technique offer scalability. Further performance improvements in DT-based digital image watermarking algorithms and DCT-based watermarking algorithms could be obtained by combining DT with DCT [9-11]. The idea of applying two transform is based on the fact that combined transforms could compensate for the drawbacks of each other, resulting in effective watermarking. In nature image, the energy of each block is concentrated on the low frequency after transformation [8]. It is known that embedding watermark in low frequency makes the watermark perceptible. On the other hand, to survive lossy data compression, watermark information should not be inserted into the higher frequency. Traditional techniques select the middle-frequency range to embed the watermark. For example, Al-Haji s [9] proposed embedding visually recognizable patterns into the images by selectively modifying the middlefrequency parts of the image. However, in our proposed method, watermark is embedded into the low frequency band of the DCT block of selected DT sub-band. To achieve perceptual invisibility of the watermark, the weighted correction which is an approach to justify the watermarked image on the spatial domain is proposed. Binary bits of watermark are embedded into the low frequency of DCT coefficients of the selected frequency subband of DT. For embedding we use two uncorrelated pseudorandom sequences. One sequence is used to embed watermark bit one and another is used to embed watermark bit zero. The algorithm for watermark extraction is reversing the embedding operations without the reference of the original image. Section II describes the proposed framework for watermarking. The performances are evaluated in Section III. Finally, the conclusion is given in Section IV. II. THE PROPOSED FRAMEORK In this section we will discuss about the watermark bits embedding process and extraction the watermark from the watermarked image. A. Embedding Approach In Figure 3. the proposed framework for embedding watermark bits into the host image is shown. To embed watermark bits in the host image the following steps are needed. 978-1-4673-1773-3/1/$31.00 013 IEEE
1) DT transform: Apply single level of DT on the host image to decompose it into four non-overlapping multiresolution coefficient sets. Figure 1. shows the four non overlapping coefficient sets of an image. The coefficient sets are given below: = x) 1 (u x)(v = ( ) ( ) 1 LH g x h y (u x)(v = (1) 1 ( ) ( ) 1 ( )( ) HL h x g y u x v y = h( x) h( 1 (u x)(v HH where is the level of the -D DT, g (n) and h (n) are the impulse responses of the low-pass and high-pass filters respectively, and 0 = ( u, is the original image. Choose HL coefficients set for embedding watermark bits. Figure 1. Single level DT decomposition of an image. ) Divide the horizontal coefficients set into 4X4 blocks: Divide the chosen coefficients set HL into 4x4 blocks. A 4 4 block of chosen coefficient set is shown in Figure. Figure. Divide the horizontal coefficients set into 4X4 blocks. 3) DCT transform on selected coefficient set: Apply DCT on each block of the chosen coefficient set HL.The DCT can be defined by the following equation: = F ( u, x, y, u, () where the kernel is given by following equation: (x + 1) uπ (y + 1) vπ x, y, u, = α ( u) α( cos cos (3) 1 whereα ( u) = α ( = for u, v = 0 Figure 3. The proposed framework for embedding watermark bits into the host image. 4) Convert watermark into binary format: Convert the watermark image into binary format. 5) Generate pseudorandom sequences: Generate two uncorrelated pseudorandom sequences by using a key. One sequence is used to embed the watermark bit 0 ( seq _ 0 ) and the other sequence is used to embed the watermark bit 1 (seq _1). Pseudorandom sequences are used to increase security level. 6) Low frequency embedding: Embed the two pseudorandom sequences with a gain factor α in the DCT transformed 4 4 blocks of the selected DT coefficient sets of the host image. Instead of embedding in all coefficients of the DCT block, it embeds only to the low frequency DCT coefficients. The DCT coefficients are stored according to a zigzag format, as shown in Figure 4. C( i, j) indicates the embedding position of the low frequency. If we denote X as the matrix of the low frequency coefficients of the DCT transformed block, then embedding is done as following equation. If watermark bit is 0, then X = X + (α seq_0). (4) otherwise α ( u) = α ( = If watermark bit is 1, then X = X + (α seq_1). (5)
Figure 4. The embedding position of the low frequency. 7) Inverse DCT: Perform the inverse DCT on each block after its low-band coefficients have been modified to embed the watermark bits as described in the previous step. 8) Inverse DT: Perform the inverse DT on the DT transformed image, including the modified coefficient set, to produce the watermarked host image. 9) ustify the watermarked image: Compare the watermarked image with original image. Let D be the difference in gray levels between the original image I 1 and the watermarked image I, it can be represented as: D = I1(i,j) I(i,j), 0 i < 1, 0 j <. (6) Let M denotes the magnitude suppression of D, that is M i, j = Di, j c. (7) The constant value of c is used to improve the invisibility of watermark. Then the watermarked image I is modified as: I i, j = Ii, j + M i, j (8) where I is the final watermarked image after weighted correction. B. Extracting approach In Figure 5. the proposed framework for extracting watermark from the watermarked image is shown. To extract watermark bits from the watermarked image the following steps are needed. 1) Filtering operation: Perform pre-filtered on the watermarked image by the combination of sharping and Laplassian of Gaussian filter to increase distinction between host image and watermarked image ) DT transform: Apply single level DT on the prefiltered watermarked image to decompose it into four nonoverlapping multi-resolution coefficient sets. Choose HL coefficients set. 3) Divide the horizontal coefficients set into 4X4 blocks: Divide the chosen coefficient sets into 4 x 4 blocks. 4) DCT transform on selected coefficient set: Apply DCT on each block in the chosen coefficient set. 5) Generate pseudorandom sequences: Regenerate the two pseudorandom sequences seq _ 0 and seq _ 1 using the same key that was used in the watermark embedding procedure. 6) Calculate correlation: For each block in the coefficient set HL calculates the correlation between the low-band coefficients and the two generated pseudorandom sequences. If the correlation with the seq _ 0 is higher than the correlation with seq _ 1, then the extracted watermark bit is considered 0, otherwise the extracted watermark is considered 1. Figure 5. The proposed framework for extracting watermark from the watermarked image. 7) Reconstruct extracted watermark: The watermark image is reconstructed using the extracted watermark bits. 8) Compute the similarity: Compute the similarity between the original and extracted watermarks. III. PERFORMACE EVALUATIO Among various test images employed in experiments, the 51 51 Lena image which is shown in Figure 6. is used to show the effectiveness of the proposed method. The 64 64 watermark is shown in Figure 6.. Figure 6. Lena image which worked as the host image atermark image which is embedded to the host image. Imperceptibility: Imperceptibility means that the perceived quality of the host image should not be distorted by the presence of the watermark. As a measure of the quality of a watermarked image, the peak signal to noise ratio is typically used. The PSR has been utilized to calculate similarity between the original image and the watermarked image. The mathematical representation is given below: 55 PSR = 10 log10 (9) MSE [ I i j I i j 1(, ) (, )] i = 0 j = 0 where MSE = (10) The MSE (mean square error) will be compute firstly and then the value for PSR will be available secondly. Here I1( i, j) and I ( m, n) respectively represent the gray value of the original image and the watermarked image. Robustness: Robustness is a measure of the immunity of the watermark against attempt to remove it, intentionally or unintentionally, by different types of attacks. e measure the similarity between the original watermark and the watermark extracted from the attacked image using the ormalized Correlation (C) factor, which is given below:
K 1 ( j, k ) ( j, k ) j = 0 k = 0 (11) C = K, ) 1 ( j k j = 0 k = 0 here 1 ( j, k) is the original watermark image and ( j, k) is the extracted watermark image. Figure 7. shows the watermarked Lena image and the PSR of the watermarked Lena image is about 35.634dB. Figure 7. shows the recovered watermark with correlation 1. Some data are provided in TABLE II with the watermarked image I is attacked by various kinds of noise. TABLE II. DIFFERET TYPES OISE ATTACKED DATA Salt and Pepper noise Gaussian noise Speckle noise Strength=0.01 Average=0, Variance=0.01 Variance=0.00 PSR=5.9093 PSR= 7.6189 PSR=7.363 C=0.9473 C=0.9878 C=0.9841 C. Filtering The watermarked image I in Figure 7. is transformed into Figure 10. after einer filtering, Figure 10. is the Extracted image from Figure 10. where correlation is 0.8900. Figure 7. atermarked image which is found after embedding watermark Extracted watermark from watermarked image. A. PEG Compression The watermarked image I in Figure 7. is transformed into a new image after PEG compression in which the quality factor is 50 and it is shown in Figure 8.. The image watermark extracted from it is shown in Figure 8.. Figure 10. atermarked image after einer filtering Extracted watermark from filtered watermarked image. D. Cropping Cropping refers to the removal of the any parts of an image. Figure 11. is shown 18 18 cropped of Figure 11.. Figure 11. is the extracted watermark. Figure 8. PEG Compression of atermarked image Extracted watermark from compressed watermarked image. In TABLE I some data are given for PEG compression. TABLE I. EPG COMPRESSIO DATA Quality factor PSR C 90 3.904 1 80 3.853 1 70 3.8063 1 50 9.1601 0.9650 B. oise attacking The watermarked image I in Figure 7. is transformed into Figure 9. after adding Gaussian noise, Figure 9. is the watermark image extracted from Figure 9.. Figure 11. atermarked image after cropped Extracted watermark from cropped watermarked image. The experimental result of proposed method is given in TABLE III after the watermarked image is cropped. TABLE III. CROPPIG DATA Cropped area PSR C 3 3 9.845 0.9839 64 64 5.6530 0.960 80 80.9744 0.940 18 18 18.5870 0.898 Figure 9. atermarked image after adding Gaussian noise Extracted watermark from Gaussian noise attacked watermarked image. E. Contrast Adjustment The watermarked image I in Figure 7. is transformed into Figure 1. after Contrast Adjustment and Figure 1. is the extracted watermark from Figure 1.. The correlation of extracted watermark with original watermark is 0.9983.
Figure 1. atermarked image after Contrast Adjustment Extracted watermark from attacked watermarked image. IV. COCLUSIOS In this article, a joint DT and DCT based watermarking technique with low frequency watermarking with weighted correction has been proposed. The technique is based on the frequency domain that the watermark is mainly inserted into the low frequency. To increase the imperceptibility, the watermark image is adjusted by the weighted correction in the spatial domain. The watermarks are embedded into different position of the low frequency for each block of selected subband of DT domain. The result of experiments have showed that the algorithm has better visibility and has stronger robustness when it is attacked by joint photographic experts group (PEG) compression, cropping, contrast adjustments, filtering, noise and so on. The experimental result shows that in most of the case the correlation between the original watermark and the extracted watermark is above than 0.9. These results demonstrate that the proposed method is suitable candidate for image copyright by which it may be possible to avoid unauthorized replication problem. REFERECES [1] Ming-Shing Hsieh, Perceptual Copyright Protection using multiresolution avelet Based atermarking and Fuzzy Logic, International ournal of Artificial Intelligence & Applications (IAIA), VOL.1, O.3, uly 010. [].ikolaidis and I.Pitas, Robust image watermarking in the spatial domain, Signal Processing Vo1.66, pp.385-403, 1998. [3] Da-Chun u and en-hsiang Tasi, Image Hiding in Spatial Domain Using An Image Differencing Approach, Proc. Of 1998 orkshop on Computer Vision, Graphics, and Image Processing, Taipei, Taiwan, pp. 80-87, 1998. [4] R. olfgang and E.. Delp, A atermark for Digital Image, IEEE int. Conf On Image Processing, Vol. 111, pp 19- Lausanne, Switzerland, September 1996. [5] Chiou-Ting Hsu and a-ling u, Multiresolution atermarking for Digital Images, IEEE Trans. Circuits and System II, Vol. 45, o. 8, pp. 1097-1101, August 1998. [6] Athanasios ikolaidis and Ioannis Pitas, Asymptotically Optimal Detection for Additive atermarking in the DCT and DT Domains, IEEE Transaction on Image Processing, Vol. 1, o. 5, May 003. [7] ai C. Chu, DCT-Based Imageatermarking Using Subsampling, IEEE Transaction on Multimedia, Vol. 5, o. 1, March 003. [8] Shinfeng D. Lin and Chin-Feng Chen, A Robust DCT-Based atermarking for Copyright Protection, IEEE Transactions on Consumer Electronics, Vol. 46, o. 3, AUGUST 000. [9] Ali Al-Haj, Combined DT-DCT Digital Image atermarking, ournal of Computer Science 3 (9): 740-746, 007. [10] Shital Gupta, Dr Sanjeev ain, A Robust Algorithm of Digital Image atermarking Based on Discrete avelet Transform, Special Issue of ICCT Vol.1 Issue, 3, 4; 010 for International Conference [ACCTA- 010], 3-5 August 010. [11] Saeed K. Amirgholipour and Ahmad R. aghsh-ilchi, Robust Digital Image atermarking Based on oint DT-DCT, Internationl ournal of Digital Content Technology and its Applications, Vol. 3, o., une 009.