International Journal of Electronics and Computer Science Engineering 802 Available Online at www.ijecse.org ISSN: 2277-1956 COMPRESSION OF 3D MEDICAL IMAGE USING EDGE PRESERVATION TECHNIQUE Alagendran.B 1, S. Manimurugan 2, M. John Justin 3 1,3 PG Research Scholar, Image Processing Group, Karunya University, Coimbatore, India 2 Assistant Professor, Dept of Computer Science and Technology, Karunya University, Coimbatore, India Email- 1 alagendran.20@gmail.com, 2 johnjuztin@gmail.com Abstract- This paper presents performing edge preservation technique in compressed 3D medical image.the 3D image compression is performed by wavelet transforms such as daubechies 4, daubechies 6, cohen daubechies feauveau 9/7, cohen daubechies feauveau 5/3. The performance parameters to calculate the 3D medical image compression are PSNR, Bit Rate. The input images used in this experiment are MRI. As a result, proposed system enhances the compressed 3D medical image quality using Edge Preservation Technique which also increases the PSNR value results in quality of the image will be improved. Keywords: 3D Medical image compression, 3D Wavelet Transform, edge preservation technique, 3D encoder I- Introduction In medical application, storing the large amount of medical data is not a easy way. Hence, the compression technique plays a vital role in storing the large amount of medical image. While compressing the medical image it performs both lossless and lossy compression. In lossless compression no data will be lost whereas in lossy compression certain amount of data may be lost. Dealing with medical images it is necessary to go with the lossless compression. To improve the clarity of the compressed image, introducing a edge preservation technique to sharpen the edges of the images promotes good quality level in output image. 2-Literature Survey N. Sriraam and R. Shyamsunder has put forth a method for 3D medical image compression using 3-D wavelet coders. In this paper the wavelet transforms were used for compressing the 3D medical image such as Daubechies 4, Daubechies 6, Cohen Daubechies Feauveau 9/7and Cohen Daubechies Feauveau 5/3 along with the encoders namely,3d SPIHT(3D set partition in heirarical trees), 3D SPECK(3D set partition in embedded block) for the further process of compression. in order to find the best pair of wavelet and the encoder. Both the symmetric and decoupled wavelet transform were used. The images used to perform the experiment such as MRI (Magnetic Resonance Images) and XA (X-ray Angiograms). The performance parameter used to evaluate the algorithm is PSNR (Peak Signal-to Noise Ratio) and bit rate. Also structural similarity is evaluated to find the difference between the original and the reconstructed images. As the result of this,the 3D Cohen Daubechies Feauveau 9/7 symmetric wavelet in combination with the 3D SPIHT encoder shows the good compression result.[1] Zixiang Xiong has proposed a Lossy to Lossless Compression method of Medical Volumetric Data Using 3-D Integer Wavelet Transforms. In this paper the 3D integer wavelet transform is used to perform the bit shifting of the wavelet coefficients for a 3D unitary transform. Also context modeling is performed for arithmetic coding of these wavelet coefficients.3dspiht and 3D subband coder with truncation were used to compress the medical volumetric data. Resulted in good range of compression both in lossy and lossless.[2] Compression and the Decompression Strategies for a Large Volume of Medical Images were proposed by Karthik Krishnana et al. In this paper compressed interactive rendering of large volume of data in distributed environments were used to promote the distortion scalability and the multi-resolution functions especially for the low bandwidth networks. The interactive client has the breadth according to the scalability in resolution, position and the quality. The client now finds the VOI (volume of interest).contextual background data is used to find the quality fading away from the VOI. Rendering process has been performed, when the client approaches the desired quality threshold value. Performance parameters used to evaluate the algorithm are compression ratio, PSNR, decode time and the data transmission.[3]
803 R. Srikanth and A. G. Ramakrishnan have designed a contextual encoding technique in the uniform and adaptive mesh based lossless compression of MR images. In this paper the number of improvements to the mesh based coding method for the 3D brain magnetic resonance images were evaluated, by eliminating the clinically irrelevant background leads to meshing of the brain part alone of the image. Also by generating the content based mesh with spatial edges and the optical flows in between the two continuous slices. Also the context based entropy coding of the residues after the motion compensation by the affine transformations. In considering the performance parameter compares the bit rates of the uniform and adaptive mesh based methods. Results showed that the adaptive mesh based method gives good results than the uniform mesh based scheme, at the rate of high complexity.[4] Peter Schelkens has proposed a technique called the wavelet coding of volumetric medical datasets. In this paper layered zero coding and context based arithmetic coding were used along with the 3D DCT (Discreet Cosine Transform) based coding for benchmarking. This coding algorithm gives the embedded data stream which is further decoded till the lossless level. Resulted in, good range of loss and lossless compression. [5] Input Image Split into group of slices 3D Wavelet Transform 3D Encoder Compressed Image Figure 1 Sriram et al model of 3D medical image compression [1] Qingquan LI, Qingwu HU has proposed a 3D wavelet compression for multiple band remote sensing images using the edge reservation technique. In this paper, a 3D wavelet analysis was done for multiple band image compression and a fast 3D wavelet transform method was proposed. For fetching the local edge features of the image, and the multiple band image compression method using 3D wavelet was used that is on the basis of contour and edge characteristics of multi band images in the same boundary. Also the remove correlation technique of holding the characteristics of contour is used with the 3D wavelet analysis. And the relative quantification coding method is used for edge preservation of multiple band images. Resulted in high compression ratio after preserving the edges of the compressed images.[6] Amir Yavariabdi, Chafik Samir, Adrien Bartoli has designed a wavelet transform for the 3D medical image edge preservation.in this paper the 2D and the 3D wavelet medical image resolution algorithm was proposed which is on the basis of interpolation of the low resolution input image and high frequency subband images by using the DWT (DiscreteWavelet Transform). results shows better performance after preserving the edges.[7] 3.1 Applying 3D Wavelet Transform 3-Related Works It is known that 3-D wavelet transform can be obtained by applying 1-D wavelet transform along each dimension.if Wx,Wy,Wz represent the wavelet transformations applied along the x, y and z axes then transform of this image is defined as W = (WxWyWz)1(WxWyWz)2... (WxWyWz)L where L is the number of levels. This is called the 3-D symmetrical wavelet transformation, where all the dimensions are decomposed to equal number of levels by applying wavelet transform alternately to each axis. On the other hand, 3-D wavelet transform can be also be obtained by applying 2-D spatial transform first, followed by wavelet transform on the z axis separately. This can be defined as,
IJECSE,Volume1,Number 2 Alagendran.B et al. W = (WxWy)1(WxWy)2... (WxWy)L(Wz)1(Wz)2... (Wz)l. The spatial axis are decomposed to equal number of levels L but the z axis can be decomposed to l levels which need not be equal to L. This is called the decoupled 3-D wavelet transform or the 3-D wavelet packet transform. The lifting approach used for implementing wavelets has been proposed by Sweldens Figure 2 The 3-D medical image data considered in this work are MR and XA image [1] The major advantage of lifting is the fact that it requires lesser memory and is faster than the general wavelet transform, hence providing a platform for real time applications. 4-Proposed system Once the medical image has been compressed it holds the blurred edges hence it is necessary to preserve the edges of the image since it gives good clarity of image. The block diagram of the proposed system is illustrated in fig 3 and the edge preservation block is explained with neat flowchart in fig 4. First the 3-D medical image is given as input image. This is further undergoing the process of splitting into groups of slices (GOS). Then the 3-D wavelet transform is applied, followed by 3-D encoding. To encode the image and to get the final compressed image, both the wavelet transforms such as symmetric and decoupled wavelet transform are used.
805 Input Medical Image Splitting into group of slices Applying 3D Wavelet Transform Performing 3D Encoder (SPIHT) Compressed 3D Medical Image Implementing Edge Preservation Technique Edge preserved compressed Medical Image Figure 3.Flowchart of Proposed System 3-D Set Partitioning In Hierarchical Trees has the following advantages High PSNR values Hierarchical Trees are Set Partitioned Superior to JPEG in image quality and PSNR value Keeps spatial oriented tree structure It produce best results in still image coding Performs measurably and visually better than JPEG Employs complicated means of motion estimation and compensation Fully embedded wavelet coding algorithm with precise rate control Low complexity 4.1 Edge Preservation Block Within the Edge Preservation Block, the process given in the following flowchart is carried out at first; the edges from the images are extracted when its frequency is high. Low frequency coefficients from the edges are taken in the other hand.
IJECSE,Volume1,Number 2 Alagendran.B et al. 3D DWT has been chosen to preserve the edges. In the proposed algorithm, one level DWT is applied to decompose a 3D low resolution image into eight different sub-band images. Edge Extraction Remove edge from f H (x, y) Band low frequency coefficient High Band keep high frequency coefficient width Edge Huffman Code Quantification Compressed Data of low Band Compressed Data of High Band Figure 4.Flowchart of Edge Preservation Block The high frequency sub-bands such as HHH, HHL, HLH, LHH, LHL, LLH, and HLL (where H and L are High and Low coefficients) contain the edges of the low resolution image. The high frequency coefficients undergo the process of quantification in order to compress the image whereas the low frequency coefficients undergo the Huffman coding process for compressing the image. Thus, the edges of the images are corrected and clear visual effects are obtained.the PSNR results in enhancement work have sharper edge features, more details, and visually it is closer to the original image when compared to the base paper PSNR results. 5-Performance parameter The performances of the 3-D medical images are evaluated in accordance with the PSNR (peak signal to noise ratio), MSSIM (Mean Structural Similarity Index) and bit rate. The performance of the wavelet based algorithms yields better results than JPEG and other 3-D transforms. MSSIM index is an image quality assessment parameter based very much on the characteristics of HVS and measures the structural similarity rather than error visibility between two images. The SSIM is defined as, SSIM (i, i ) = (2µiµi + C1) + 2σii + C2 (1) Where, (µ2i µ2i + C1)(σ2i σ2i + C2) i and i are spatial patches (windows) of original image I and reconstructed image I respectively, µi and µi are the mean intensity values of i and i respectively,
807 σi and σi are standard deviation of i and i respectively, and C1, C2 are constants. The mean value is given by, MSSIM (I, I ) = 1/M SSIM (i i (2) Where, M is the number of windows of the image. The PSNR value is calculated using, PSNR (db) = 20 log (Maximum pixel value) / MSE (3) Where, MSE represents the mean squared error of the image defined as, MSE = 1/N fi,jfi,j (4) Where, N is the total number of pixels, F (i, j) denotes the pixel value in the reconstructed image and f (i, j) is the pixel value in the original image. BR (bpp) = Size of the compressed image in bits (5) Total no of pixel If the bit rate increases, it results in improvement in quality of the reconstructed image. The performance of the wavelet based algorithms yields better results than JPEG and other 3-D transforms. It can be concluded that the user can fix the bit rate depending on his reconstructed image quality requirements. 6. Results and discussions The experiment result is analyzed while compressing the 3D medical image. The 3D medical image compression using Edge Preservation Technique generates high PSNR values and improves the quality of the image which performs better results than the compressing the 3D medical image compression without Edge Preservation Technique. The following Table1,Table2 depicts the PSNR Value of the compressed 3D medical Image without edge preservation technique and with edge preservation technique respectively. Also Fig 5 represents the comparison chart of the PSNR value using with and without Edge Preservation Technique. Table 1.3D Medical Image Compression without Edge Preservation Technique Bit Rate 0.5 32 1 38 1.5 48 2 56 2.5 64 PSNR Values
IJECSE,Volume1,Number 2 Alagendran.B et al. Table 2.3D Medical Image Compression with Edge Preservation Technique Bit Rate PSNR Values 0.5 38 1 43 1.5 59 2 67 2.5 74 100 80 60 40 20 0 1 2 3 4 5 6 3d_SPIHT 3D-SPIHT WITH EDGE PRESERVATION Figure 5.Comparison chart of 3D Medical Image Compression, without Edge Preservation Technique and with Edge Preservation Technique 7-Conclusion This paper explains about the improving the compressed 3D medical images using edge preservation technique. The proposed technique applies a shape function to the input low resolution image in order to enhance the discontinuities. Comparisons based on PSNR and visual results demonstrate that the proposed method provides the best result in terms of PSNR. To conclude the experimental results of compressed 3D medical image with edge preservation technique yields better results than compressed 3D medical image without edge preservation. References [1] N. Sriraam, R. Shyamsunder. 3-D medical image compression using 3-D wavelet coders, Elsevier on Digital Image Processing,Vol.21,pp.100-109,2010. [2] Zixiang Xiong, Xiaolin Wu, Samuel Cheng and Jianping Hua, Lossy-to-Lossless Compression of Medical Volumetric Data Using Three- Dimensional Integer Wavelet Transforms, IEEE Transactions On Medical Imaging, Vol. 22, No. 3, pp.459-470, 2003. [3] Karthik Krishnan, Michael W. Marcellin, Ali Bilgin, Mariappan Nadar, Compression Decompression Strategies for Large Volume Medical Imagery, Medical Imaging: PACS and Imaging Informatics, Vol. 5371,pp 152 159,2004. [4] R. Srikanth and A. G. Ramakrishnan, Contextual Encoding in Uniform and Adaptive Mesh-Based Lossless Compression of MR Images, IEEE Transactions On Medical Imaging, Vol. 24, No. 9, pp.1199-1206, 2005. [5] Peter Schelkens, Joeri Barbarien, Mihnea Galca, Xavier Giro-Nieto and Jan Cornelis, Wavelet Coding of Volumetric Medical Datasets, IEEE Transactions On Medical Imaging, Vol. 22, No. 3, pp.441-458,2003. [6] Qingquan LI, Qingwu HU, 3D Wavelet Compression To Multiple Band Remote Sensing Images Based On Edge Reservation, Spatial Information and Network Communication Research and Development Center, Wuhan University. [7] Amir Yavariabdi, Chafik Samir, Adrien Bartoli, 3D Medical Image Enhancement based on Wavelet Transforms, ALCoV-ISIT Université d Auvergne Clermont-Ferrand,pp.1-5,2011 [8] J. Jyotheswar, Sudipta Mahapatra, Efficient FPGA implementation of DWT and modified SPIHT for lossless image compression, Journal of Systems Architecture, Vol.53, pp.369 378, 2007.
809 [9] Yen-Yu Chen, Medical image compression using DCT-based sub band decomposition and modified SPIHT data organization, International journal of medical informatics, Vol.76, pp. 717 725, 2007. [10] R. Srikanth, A.G. Ramakrishnan, Contextual encoding in uniform and adaptive mesh based lossless compression of MR images, IEEE Transactions on Medical Imaging,Vol. 24,pp.1199 1206, 2005. [11] R. Shyam Sunder, C. Eswaran, N. Sriraam, Medical image compression using 3-D Hartley transform, Computational Biological Medicine, Vol.36, pp.958 973, 2006. AUTHORS BIOGRAPHY 1. B.Alagendran received the B.E degree in Computer Science and Engineering from the Anna University, Chennai, India, in 2010, and pursuing his M.Tech degree in Software Engineering in Karunya University, Coimbatore, India. His research interests include image processing, software engineering, data mining. 2. S.Manimurugan received the B.E. degree in Computer Science and Engineering from the Anna University, Chennai, India, in 2005, and the M.E. degree in Computer Science and Engineering in 2007. He is currently pursuing the Ph.D. degree in Computer Science and Engineering in Anna University, Coimbatore, India. His current research interests are in Image Processing, Information Security. 3. M.John Justin received the B.E degree in Computer Science and Engineering from the Anna University, Chennai, India, in 2010,and pursuing his M.Tech degree in Software Engineering in Karunya University, Coimbatore,India.His research interests include image processing, software engineering.