126 Int. J. Medical Engineering and Informatics, Vol. 5, No. 2, 2013 DICOM medical image watermarking of ECG signals using EZW algorithm A. Kannammal* and S. Subha Rani ECE Department, PSG College of Technology, Peelamedu, Coimbatore, 641004, Tamilnadu, India Fax: 00-91-0422-2572177 E-mail: kannammal_a@yahoo.com E-mail: ssrani61@yahoo.co.in *Corresponding author Abstract: A novel blind watermarking method with secret key is proposed by embedding ECG signals in medical images combined with the EZW-based wavelet coder. At first, the host image and the mark signal are decomposed using discrete wavelet transform. The image starts to be encoded using EZW algorithm for its maximum threshold. Then threshold is halved and encoding process is repeated. Watermark insertion begins when the signal threshold is reached. After entropy coding of the bits to be transmitted, the watermarked image is progressively encoded. At the receiver both the inverse EZW algorithm is applied and both the image and the signal are reconstructed. The watermarking is done for 128 to 8,192 bytes of the mark signal. The peak signal to noise ratio (PSNR) between the original and watermarked image is found to be as high as 80 db and increases as the size of the watermark decreases. Keywords: DICOM; EZW; BPP; discrete wavelet transform; DWT; ECG; watermark. Reference to this paper should be made as follows: Kannammal, A. and Subha Rani, S. (2013) DICOM medical image watermarking of ECG signals using EZW algorithm, Int. J. Medical Engineering and Informatics, Vol. 5, No. 2, pp.126 136. Biographical notes: A. Kannammal received her PhD in Engineering from Anna University, Coimbatore in 2008, ME from Thiagarajar College of Engineering in 2004, and BE in Electronics and Communication Engineering in 2002. In 2004, she joined the Department of ECE, Thiagarajar College of Engineering as a Lecturer. She joined PSG College of Technology as a Lecturer in the Department of ECE in 2007. She is currently working as an Assistant Professor in the Department of ECE at PSG College of Technology. Her interests include medical image processing and soft computing techniques. She has contributed six technical papers in various journals. S. Subha Rani received her PhD in Engineering from Bharathiar University in 2000, ME from Anna University in 1987, and BE in Electronics and Communication Engineering in 1984. In 1988, she joined the Department of ECE as a Lecturer. She is currently working as Professor and Head in the Department of Electronics and Communication Engineering at PSG College of Technology, Coimbatore. Her interests include optimal control, wireless Copyright 2013 Inderscience Enterprises Ltd.
DICOM medical image watermarking of ECG signals using EZW algorithm 127 systems, MEMS and soft computing techniques. She has contributed more than 67 technical papers in various journals and conference. She is a life member of IEEE and IETE. She has completed four sponsored projects. 1 Introduction The demand for authentication of medical images is enormous, due to the increasing needs of data exchange in inter or intra hospital network for diagnosis and with external parties such as research institutes. Exchange of database between hospitals needs efficient and reliable transmission and storage techniques to cut down the cost of healthcare. This exchange involves large amount of vital patient information such as bio-signals and medical images. Interleaving one form of data such as 1-D signal, over digital images can combine the advantages of data security with efficient memory utilisation. Embedding vital information of patients inside their scan images will help physicists to make a better diagnosis of disease. A novel blind watermarking method with secret key by embedding ECG signals in medical images combined with the EZW-based wavelet coder. The principle is to replace significant wavelet coefficients of the host image by the corresponding significant wavelet coefficients belonging to the ECG signal. At first, the host image and the mark signal are decomposed using discrete wavelet transform (DWT). The image starts to be encoded using EZW algorithm for its maximum threshold. Then threshold is halved and encoding process is repeated. Watermark insertion begins when the signal threshold is reached. After entropy coding of the bits to be transmitted, the watermarked image is progressively encoded. At the receiver the inverse EZW algorithm and inverse DWT are applied and both the image and the signal are reconstructed (Nambakhsh et al., 2006). Figure 1 Flowchart of the proposed algorithm 2 EZW in image 2.1 2-D discrete wavelet transform The DWT is a recurrent process of low and high pass filtering performed on the input dataset. In the first iteration, the whole data vector is low and high-pass filtered,
128 A. Kannammal and S. Subha Rani and the resulting vectors are stored separately. This means that every second element in the vectors is removed. The new high pass filtered vector is now called a detail coefficient vector. The coefficients in this vector are wavelet coefficients. These two vectors form the first scale of the DWT. The second and subsequent iterations of the filtering process are performed on the vector generated by the previous iteration s low-pass filter. The vector from the high-pass filter is stored. Each iteration adds another scale to the wavelet transformed data (Nambakhsh et al., 2006). 2.2 Overview of EZW The EZW encoder is based on the principle of progressive encoding or embedded encoding, which means that when more bits are added to the bit stream, the decoded image will contain more details. The wavelet transform is a dyadic decomposition of an image achieved by a pair of quadratic mirror filters (QMF). 2.2.1 Observations of EZW The EZW encoder is based on two important observations: Natural images in general have a low pass spectrum. When an image is wavelet transformed the energy in the sub-bands decreases as the scale decreases (low scale means high resolution), so the wavelet coefficients, be smaller in the higher sub-bands than in the lower sub-bands. This show that progressive encoding is a very natural choice for compressing wavelet transformed images, since the higher sub bands only add detail. Large wavelet coefficients are more important than small wavelet coefficients. 2.2.2 Properties of EZW The EZW algorithm exploits the hierarchy of the wavelet coefficients, and establishes a connection between coefficients from different sub-bands, allowing multiple coefficients to be encoded simultaneously. Secondly, coefficients are encoded in order of importance using bit prioritisation. 2.2.3 EZW encoding The above mentioned observations are exploited by encoding the wavelet coefficients in decreasing order, in several passes. For every pass a threshold is chosen against which all the wavelet coefficients are measured. If a wavelet coefficient is larger than the threshold it is encoded and removed from the image, if it is smaller it is left for the next pass. When all the wavelet coefficients have been visited the threshold is lowered and the image is scanned again to add more detail to the already encoded image. This process is repeated until all the wavelet coefficients have been encoded completely or maximum bit rate has been satisfied.
DICOM medical image watermarking of ECG signals using EZW algorithm 129 2.2.3.1 Quad tree After wavelet transforming an image, it is represented using trees because of the sub sampling that is performed in the transform. A coefficient in a lower sub band can be thought of as having four descendants in the next higher sub band. The four descendants each also have four descendants in the next higher sub-band and a quad-tree emerges. 2.2.3.2 Zero tree A zero tree is defined as a quad-tree of which all nodes are equal to or smaller than the root. The tree is coded with a single symbol and reconstructed by the decoder as a quad-tree filled with zeroes. Also the root has to be smaller than the threshold against which the wavelet coefficients are currently being measured. 2.2.4 EZW CODEC The EZW encoder consists mainly of two algorithms, viz., the dominant pass and the subordinate pass. The dominant pass and subordinate pass procedures are the algorithms for the EZW encoder that helps in selecting the significant values and discarding the sub threshold values. The EZW algorithm calculates the initial threshold, using the following formula. T = initial ( log 2( k ) 2 ) where T initial is the initial threshold and k is the maximum absolute coefficient value. Figure 2 Flow chart for EZW encoding
130 A. Kannammal and S. Subha Rani Figure 3 Flow chart for dominant pass 2.2.4.1 Subordinate pass The flow chart for the subordinate pass procedure is shown in Figure 4. Figure 4 Flow chart for subordinate pass
DICOM medical image watermarking of ECG signals using EZW algorithm 131 3 EZW in signals The EZW algorithm can be applied to signals similar to an image. In this case decomposed signal coefficients have dyadic tree instead of quad tree. The similarity between images and signals is used to embed the signal inside an image. Figure 5 Dyadic tree of 1-D signals (see online version for colours) 4 Proposed algorithm 4.1 Watermark insertion process Five levels of wavelet decompositions are performed using Db1 as mother wavelet to obtain matrices of wavelet coefficients of host image and signal. Then the maximum threshold of image and signal are calculated. In the encoding process of the EZW coder the image starts to be coded. After this, it is looked for the dominant pass with a threshold same as the maximum threshold of the signal. The coefficients of the image belonging to this interval are replaced by the signal coefficients belonging to the same interval according to a predefined path. 4.2 Watermark extraction process In the extraction process most of the steps of the insertion process are repeated but in the reverse order, and the wavelet coefficients of the compressed watermarked image are constructed. Then, scanning is performed on the matrix in a predefined way using the secret keys to find signal coefficients. In this step, only the wavelet coefficient matrix of the watermarked compressed image is needed to construct the extracted wavelet coefficients of the ECG signal. By using the secret key the signal coefficients that are embedded in the image are found. Finally, the matrix of wavelet coefficients of the ECG is constructed, and then 1-D and 2-D inverse discrete wavelet transform (IDWT) is applied to the matrix in order to reconstruct the extracted watermark and the compressed watermarked image respectively.
132 A. Kannammal and S. Subha Rani 5 Results and discussion MRI and CT DICOM images of size 256 256 and 512 512 pixels were watermarked by embedding watermarks of size 128 bytes to 8,192 bytes. The DICOM images were obtained from KMCH and image archive. The ECG signals were obtained from the biomedical engineering department. The simulation platform used was MATLAB 7.4. The simulation results were tabulated in Table 1. The peak signal to noise ratio (PSNR) calculated between the original image and the watermarked image was found to be close to and greater than 80 db for all images as shown in the tabular columns. The PSNR (Kobayashi et al., 2009) increases as the size of the watermark increases and this is well depicted through Figure 13. The proposed method of selecting the insertion sites ensures both watermarking (Piao et al., 2008) constraints: imperceptibility, since the watermark coefficients of highest scale are almost identical or even equal to those that they are been replaced with, and robustness since the substituted coefficients are significant, and they will not be lost in the quantisation step. Blindness is obtained by the fact that only the secret key and the header information are necessary to extract the watermark (Giakoumaki et al., 2006). 5.1 Results for mr-mono2-16-knee image and brain cross sectional view Figure 6 Input host image, (a) MR-MONO2-16-KNEE (b) Brain cross sectional view (c) Wavelet decomposed image of (a) (d) Wavelet decomposed image of (b) (a) (b) (c) (d) Figure 7 ECG signal
DICOM medical image watermarking of ECG signals using EZW algorithm 133 Figure 8 Decomposed ECG signal Figure 9 Reconstructed host image Figure 10 Reconstructed ECG signal
134 A. Kannammal and S. Subha Rani Table 1 Output values for varies images Sl. no. Image type Image size Mark signal size (in bytes) PSNR (in db) 1 CT-MONO2-16-brain 512 512 128 91.3021 5.1310 256 91.2964 512 91.2082 1,024 91.2490 2,048 91.0309 4,096 90.0632 8,192 89.6561 2 CT-MONO2-16-ankle 256 256 128 81.4183 3.0288 256 81.4210 512 81.4063 1,024 81.3912 2,048 81.3650 4,096 81.2749 8,192 81.1004 3 CT-MONO2-16-chest 512 512 128 95.6906 4.4954 256 95.4808 512 95.5168 1,024 95.5056 2,048 93.9264 4,096 92.5308 8,192 90.9598 4 CT-MONO2-16-ort 512 512 128 78.5954 6.4501 256 78.5963 512 78.5934 1,024 78.5917 2,048 78.5716 4,096 78.5034 8,192 78.4282 5 Brain cross sectional view 512 512 128 84.3916 5.0626 256 84.3927 512 84.3853 1,024 84.3835 2,048 84.3262 4,096 84.3209 8,192 84.1443 BPP
DICOM medical image watermarking of ECG signals using EZW algorithm 135 Table 1 Output values for varies images (continued) Sl. no. Image type Image size Mark signal size (in bytes) PSNR (in db) 6 Brain rear view 256 x 256 128 88.4427 4.1824 256 88.4561 512 88.4258 1,024 88.3685 2,048 88.2813 4,096 87.9998 8,192 87.4284 7 MR-MONO2-16-knee 256 256 128 92.9239 6.3224 256 93.1531 512 92.7959 1,024 92.8173 2,048 91.0348 4,096 90.0200 8,192 89.5069 Figure 11 Reconstructed host image BPP Figure 12 Reconstructed ECG signal
136 A. Kannammal and S. Subha Rani Figure 13 Comparison of PSNR (see online version for colours) 6 Conclusions The proposed watermarking algorithm is blind because it only needs an initial key in the extraction process. Also this is a resolution controlled watermarking algorithm. Real time DICOM images were used for processing. From the PSNR results obtained we can infer that the watermarked images are perceptually or visually acceptable. The algorithm has also proved its imperceptibility since the embedding process here means a substitution of original coefficients by the equivalents, in the sense of EZW significance, from the watermark. This is a very significant advantage from a diagnostic point of view since the original image and the watermarked image almost look alike. The novelty brought about in this algorithm is that the ECG signal sent as a watermark is fully reconstructed. It solves its purpose as a mark signal providing authentication to the image as well as it can be used in aiding diagnostics. In inter or intra hospital network data exchange, if the medical image as well as the ECG signal is available as a single authenticated entity, efficient memory utilisation is obtained along with data security. References Giakoumaki, A., Pavlopoulos, S. and Koutsouris, D. (2006) Secure and efficient health data management through multiple watermarking on medical images, Med. Bio. Eng. Comput., August, Vol. 44, No. 8, pp.619 631. Kobayashi, L.O.M., Furuie, S.S. and Barreto, P.S.L.M. (2009) Providing integrity and authenticity in DICOM images: a novel approach, IEEE Transactions on Information Technology in Biomedicine, July, Vol. 13, No. 4, pp.582 589. Nambakhsh, M.S., Ahmadian, A., Ghavami, M., Dilmaghani, R.S. and Karimi-Fard, S. (2006) A novel blind watermarking of ECG signals on medical images using EZW algorithm, Proceedings of the 28th IEEE EMBS Annual International Conference, New York City, USA, 30 August to 3 September. Piao, C-R., Woo, D-M., Park, D-C. and Han, S-S. (2008) Medical image authentication using hash function and integer wavelet transform, 2008 Congress on Image and Signal Processing.