Contour Extraction & Compression from Watermarked Image using Discrete Wavelet Transform & Ramer Method Ali Ukasha, Majdi Elbireki, and Mohammad Abdullah Abstract In this paper we have implemented a digital watermarking technique based on single level discrete wavelet transform (DWT). The technique a watermark bits embedded into the selected high-pass filter coefficients of a cover image by using zonal methods. This scheme requires side information (high pass filter coefficients) in watermark recovery. Experimental results show that the watermark is robust to geometric attacks (Gaussian noise is used in this paper). The contours of the original image can be extracted easily by using of SSPCE (single step parallel contour extraction) method from blurred watermarked image. To compare the results, the mean square error, signal-to-noise ratio criterions, and compression ratio (or bit per pixel) were used. Experimental results for contour extraction provide the comparative results in between these algorithms in terms of normalized correlation (NC). The simplicity of the method with accepted level of the reconstruction is the main advantage of the proposed algorithm. Keywords Digital watermarking, DWT transform, contour extraction, Ramer algorithm I. INTRODUCTION N the recent years, a huge amount of digital information is Icircuiting through all over the world by means of the World-Wide Web. Most of this data is exposed and can be easily forged or corrupted. The need for intellectual property rights protection arises. Digital watermarking has been proposed as one of the possible ways to deal with this problem, to keep information safe. The watermarking of digital data has become very popular approach for intellectual property rights protection. Several watermarking techniques were developed and a large amount of methods were proposed, but still the most of known ways to protect data are far from ideal. The digital data of the various types such as text, images, audio, and video can be processed by the watermarking procedure. Two main methods for embedding are used, namely the spatial and the frequency domain. The spatial domain techniques are more vulnerable in common image attacks such Ali Ukasha is with the Sebha University, Faculty of Engineering, Electronics Department, Brack Alshatti, P.O.Box 68, Libya (corresponding author to provide phone: +218 535 86 76; e-mail: elokshy@yahoo.com). Majdi Elbireki& Mohammad Abdullah are with the Tun Hussein Onn Malaysia (UTHM), Faculty of Electrical& Electronic Engineering, Malaysia; e-mails: majdi_elbreki@yahoo.com & faiz@uthm.edu.my). as filtering or JPEG compression. The frequency-domain approaches are the most popular for image watermarking. In these schemes, the image is being transformed via some common frequency transform and watermarking is achieved by altering the transform coefficients of the image. The transforms that are usually used are the DCT, DFT and the DWT [1], [2], [3]. The stored information in a digital format can be easily transferred without any quality losses [4]. Frequency domain watermarking techniques is more effective with respect to achieving the imperceptibility and robustness requirements of digital watermarking algorithms [5] compared with spatial domain techniques [6]. Digital watermarking technique must satisfy some properties such as it must be invisible, compatible with the host, and easily extracted in a reliable and convenient way [7]. Watermarking is a technique used to hide data or identifying information within digital multimedia. Our discussion will focus primarily on the watermarking of digital image. Digital watermarking is becoming popular, especially for adding undetectable identifying marks, such as author or copyright information. Because of this use, watermarking techniques are often evaluated based on their invisibility, recoverability, and robustness. It is often desired to retrieve the embedded information without reference to the host data; this is known as blind watermarking [8]. Our goal was to implement watermarking method and evaluate their susceptibility to attack by various image processing techniques. The simulation is done using Matlab programming to add and extract watermarks. The watermark embedding was performed in the transform domain. According to the proposed model, in the discrete wavelet largest coefficients was replaced by the linear combination of the watermark. The quality of the resulted watermarked images was measured and analyzed before and after blurred by Gaussian. Recommendations for the embedding system were stated. Watermark robustness against blurred image by Gaussian attacks was verified. Number of detection experiments with accordance to embedding parameters was made. The spectrum image (high-pass filter coefficients) was considered as a possible way to detect embedded watermark. For binary image obtained the suitable threshold is applied to the blurred watermarked image. Then the contours extraction are detection using single step parallel contour extraction (SSPCE) method [9], [10]. 106
II. DISCRETE WAVELET TRANSFORM To solve difficult problems of physics, computers, and mathematics, wavelet transform is used. It allows difficult problems to be decomposed into elementary form and then reconstructed with high precision. DWT provides various applications like compression, image processing, signal processing, etc [11], [12]. Wavelet transforms are based on small wavelets with limited duration (this paper uses Haar). The forward and inverse DWT are defined in equation (1) & equation (2) respectively. Fig. 2 Zonal method using algorithm II Where ψ a = f t ψ t t * () () f() t a ψ () t = k j Two dimensional wavelet transform can be considered as an extension of 1D wavelet transform. DWT s are particularly effective in analyzing waveforms which have spikes or pulses buried in noise. 2D- signals such as images can be decomposed using many wavelet decomposition filters in many ways (we use Haar wavelet filter). III. ZONAL SAMPLING METHODS A lot of zonal methods which was described in [13], shows that the best scheme for compression and contour extraction is as illustrated in Fig. 1. Fit criterion of the algorithm consists in selecting one of the squared block of the spectral images (e.g. shadow region) as LPF filter for image compression and the other coefficients will be taken into account in the contour reconstruction stage as shown in the Fig. 2. This algorithm in this work is referred as algorithm I [13], [14]. j ( ) /2 () t = j 2 ψ 2 t k (1) (2) IV. DESCRIPTION OF THE ALGORITHM The forward single level DWT transform is applied to the gray-level image. By using low and high-pass filters after the zonal procedures (algorithms I & I the two spectral subimages are obtained for each sub images. The HPF coefficients for details coefficients are used to embedding process. The digital water mark is the embedded to the N largest values in HPF details sub-images coefficients using the following equation New _ coefficient = ( Coefficient _ of _ HPF) * ((1 + αw ) (3) Where α is parameter determined the coefficient value and the W is watermark bits. The inverse wavelet transform is taken to the combined low and high pass filters images. Then the watermarked image is attacked geometrically. The SSPCE (single step parallel contour extraction) method is applied to the binary image which is obtained by suitable threshold value applied to the noisy digital watermarked image. Flowchart of the proposed embedded digital watermark & contour extraction and compression is depicted in Fig. 3. The extraction step of watermark from host image is similar to the process of the embedded algorithm. The watermarked image must be transformed to frequency domain by DWT approach. The N largest coefficients of the spectral image (HPF) are determined. The inverse wavelet transform is applied to extract the watermark using the formula Coefficient _ of _ HPF( Watermarked _ Im ge) X = [ 1] /α (4) Coefficient _ of HPF( Original) Where is extracted watermark bits. Fig. 1 Zonal method using algorithm I This paper compared this zonal method with a another zonal method consists in selecting one block of the spectral images (i.e. shadow region) as LPF for image compression and the other coefficients will be taken into account in the contour reconstruction stage. This algorithm is referred as algorithm II and is shown in Figure 2 [14]. Flowchart of the digital watermark extraction is depicted in Fig. 4. The analyzed algorithms use method of contour extraction called (SSPCE) with 3x3 pixels window structure. By using the central pixel the object contours is extracted and the all possible edge direction is found which connects the central pixel with one of the remaining pixels surrounding it [9], [10]. 107
Image Wavelet Zonal methods (LPF &HPF) N Largest values in HPF Key Inverse Wavelet following equations (5) & (6) for bit per pixel &compression ratio are used respectively. bpp = S *8 ( n * m) (5) Binary image SSPCE method Fig. 3. Flowchart of the embedded digital watermark & contour extraction and compression Noisy watermarked image DWT Original Digital watermark Geometric Attacks Contour extraction Zonal method (HPF) HPF coefficients Compare Threshold Watermark N Largest values in HPF Digital watermark extraction Blurred watermarked image Water- Marked Image Contour compression using Ramer algorithm IDWT Fig. 4 Flowchart of the of the digital watermark extraction As a further test, the Tools image was geometrically attacked by the Gaussian. A zero-mean Gaussian noise with standard deviation 11 was used. Though the image degradation is so heavy that it cannot be accepted in practical applications, the mark is still easily recovered. ( LCC LAC ) CR = 100% (6) L CC Where: Sis Coefficients number in the desired zonal used as LPF filter; n * mis size of the image; L CC is original contour length and L AC is approximating contour length. The mean square error (MSE) and peak signal-to-noise ratio (PSNR) criterions were used to evaluate the distortion introduced during the image reconstruction and contour compression procedures. The MSE & PSNR criterion are defined by the equations (7) & (8) respectively. 1 MSE( I, = ( n * m) n i= 0 j= 0 ( I( i, j) ( I( i, j)) Where I and I are the grey-level and reconstructed images respectively. 2 ( L 1) PSNR( I, = 10log10 MSE( I, where is the grey-level number. VII. RESULTS OF THE EXPERIMENTS To visualize the experimental results a tools image &digital watermark image are selected. Selected images are shown in Figure 5. Text of digital watermark means "SebhaUniversity" in Arabic language. To obtain blurred watermarked image we create a point-spread function, PSF, corresponding to the linear motion across some pixels (LEN), at an angle of certain degrees (THETA). Images results of blurred watermarked, extracted digital watermark and contour compression using Ramer method by using of algorithms I & II zonal are shown in Fig. 6 to Fig. 9 (related results are shown in Tables I to IV). m 2 (7) (8) V. RAMER METHOD FOR CONTOUR COMPRESSION Ramer presented an iterative method which starts with aninitial segmentation and splits the segment at the point which has the farthest distance from the corresponding segment unless the approximation error is no more than the pre-specified tolerance [15]. The vertices of an edge of the approximating polygon are determined by these stored points. a) b) Fig. 5 Test images: a) Host (Tools), and b) Watermark VI. APPLIED MEASURES To evaluate theimage & contour compression ability, the 108
TABLE I BLURRED WATERMARKED & EXTRACTED DIGITAL WATERMARK IMAGES USING ALGORITHM I (N=M=66): BIT PER PIXEL = 0.2697 Images LEN THETA Autocorrelation a) 11 5 0.9843 b) 13 7 0.8962 c) 41 18 0.7206 d) 21 11 0.6851 Contour Extraction Using SSPCE Method Fig. 7. Results of contours compression images using algorithm I (N=M=66) TABLE III BLURRED WATERMARKED & EXTRACTED DIGITAL WATERMARK IMAGES USING ALGORITHM II (N=M=47): BIT PER PIXEL = 0.2699; Images LEN THETA Autocorrelation a) 11 5 0.9943 b) 13 7 0.9618 c) 41 18 0.8066 d) 21 11 0.7529 Fig. 6. Results of extracted digtal watermark algorithm I of zonal TABLE II CONTOUR COMPRESSION USING RAMER METHOD & ZONAL SAMPLING (ALGORITHMS Method PSNR CR [%] Thresholds MSE [db] a) 0.1 0.0000 Inf 58.2377 b) 0.8 0.0559 48.9373 78.5107 c) 1.2 0.1321 44.8564 82.5763 109
Fig. 8. Results of extracted digtal watermark algorithm II of zonal Thresholds TABLE IV CONTOUR COMPRESSION USING RAMER METHOD & ZONALSAMPLING (ALGORITHMS I METHOD PSNR MSE [db] CR [%] a) 0.1 0.0000 64.5605 58.3937 b) 0.83 0.0588 49.1027 78.5300 c) 1.115 0.1204 45.8304 82.6748 Contour Extraction Using SSPCE Method Fig. 9. Results of contours compression imagesusing algorithm II (N=M=47) VIII. CONCLUSION New technique for contour compression (Ramer method used in this work) from blurred watermarked image is presented. By using single level of wavelet transform the digital watermark is embedded in selected high pass filter coefficients of detailed sub-images using zonal methods. The results show that this kind of algorithms has a satisfactory performance under image blurred by Gaussian noise. The extracted contours are obtained from blurred digital watermarked image using SSPCE contour extraction method. Simulation results using MATLAB programming show that using both algorithm of zonal method the digital watermark can be extracted with normalized autocorrelation greater than 0.7. However using second algorithm of zonal method the compression ratio of contours can be improved by about 0.9 decibels. REFERENCES [1] I. Cox, J. Kilian, F. Thompson Leighton, T. Shamoon, Secure Spread Spectrum Watermarking for Multimedia, IEEE Trans. On Image Processing, vol.6, No.12, December 1997. [2] D. Tzovaras, N. Karagiannis, M. Strintzis, Robust Image Watermarking in the Subband or Discrete Cosine Transform Domain, EUSIPCO'98,- Ninth European Signal Processing Conference, Rhodos, Greece, 1998, pp.2285-2288. [3] S. Lin, C. Chen, A robust DCT-based watermarking for copyright protection, IEEE Trans. on Consumer Electronics, Vol.46, No.3, 415 421, Aug 2000. [4] R. Ibrahim, and T. S. Kuan, Steganography Imaging (SIS): Hiding Secret Message inside an Image '. Proceedings of the World Congress on Engineering and Computer Science, 2010, San Francisco, USA. [5] L. Baisa,R. Gunjal, An overview of transform domain robust digital image watermarking algorithms, Journal of Emerging Trends in Computing and Information Sciences, 2010. [6] G. RoslineNesaKumari, B. Vijaya Kumar, L. Sumalatha, and V. Krishna, Secure and Robust Digital Watermarking on Grey Level Images, International Journal of Advanced Science and Technology, 2009. [7] D. Kunder, "Multi-resolution Digital Watermarking Algorithms and Implications for Multimedia Signals", Ph.D. thesis, university of Toronto, Canada, 2001. [8] J. Eggers, J. Su and B. Girod," Robustness of a Blind Image Watermarking Scheme", Proc. IEEE Int. Conf. on Image Proc., Vancouver, 2000. [9] A. Dziech, W. S. Besbas, Fast Algorithm for Closed Contour Extraction, Proc. of the Int. Workshop on Systems, Signals and Image Processing, Poznań, Poland, 1997, pp. 203-206. [10] W. Besbas, Contour Extraction, Processing and Recognition, Poznan University of Technology, Ph. D. Thesis, 1998. [11] C.C. Lai, C.C. Tsai, Digital Image Watermarking Using Discrete Wavelet Transform and Singular Value Decomposition, IEEE Trans. On Instrumentation and Measurement, Vol. 59, No. 11, 3060-3063, Nov 2010. [12] R. Islam, R. Sifuzzaman, Z. Ali, Applications of wavelet transform and its advantages compared to Fourier transform, Journal of physical science, 2009. [13] A. Dziech, A. Ukasha, and J. Wassermann, A New Method for Contour Extraction and Image Compression in Spectral Domain, International Conference on Multimedia, Signal processing and Communications, Zadar, Croatia,pp. 41 44 (2006). [14] A. Ukasha, An Efficient Zonal Sampling Method for Contour Extraction and Image Compression using DCT Transform, The 3rd Conference on Multimedia Computing and Systems (ICMCS'12), Tangier, Morocco, May,2012. [15] U. Ramer, An iterative procedure for the Polygonal approximation of plane curves, Computer Graphics and Image Processing, Academic Press, 1972, pp. 244-256. 110