COMPRESSION OF DICOM IMAGES BASED ON WAVELETS AND SPIHT FOR TELEMEDICINE APPLICATIONS

Transcription

1 COMPRESSION OF IMAGES BASED ON WAVELETS AND FOR TELEMEDICINE APPLICATIONS 1 B. Ramakrishnan and 2 N. Sriraam 1 Dept. of Biomedical Engg., Manipal Institute of Technology, India rama_bala@ieee.org 2 Faculty of Information Technology, Multimedia University, Malaysia ABSTRACT Digital Imaging and Communications in Medicine () is a standard used for distribution and viewing of medical images from different modalities. In this paper, we propose a method for compressing images based on the well-known Set Partitioning in Hierarchical Trees () which possess the progressive transmission capabilities useful for telemedicine applications. This is in sharp contrast to Joint Photographic Experts Group () where the images have to be first compressed to the required level and only then can be transmitted. In our method, the header and image data is separated from the image where the header is transmitted first with necessary modifications. The image data is then compressed using and transmitted. It is found from the experiment that our method gives better results in terms of compression and time. Key Words:, Compression,,, Wavelets INTRODUCTION standard has been developed due to the emergence of different medical imaging modalities, which requires some standards for the purpose of storage and transmission [1]. format has a header that contains information about the image, imaging modality and information about the patient. In addition to this, there is a field in the header which defines the Transfer Syntax Unique Identification that indicates the type of compression applied to the image data. The image data then follows the header. images are usually stored in the uncompressed raw data format. The disadvantage of this is that it increases the storage size and the bandwidth requirement for transmission. Currently, standard supports run-length coding (RLE), lossy, lossless, (lossless and near lossless) and for compression of the image data [2]. In this paper, images are compressed using a progressive transmission coder, namely,. The image header is transmitted first and the image details are then transmitted in successive stages. The receiver can terminate the transmitting data at any instant once the required information is satisfactory. To evaluate the performance, we tested the image compressed with lossy, (near lossless), and. The experiments were conducted with X-ray chest image and MRI brain image and the results are evaluated in terms of peak signal-to-noise ratio (PSNR) and mean structural similarity index (). For comparison purpose, the test images are also evaluated using compression schemes, such as,, and. WAVELET BASED COMPRESSION Wavelets are used for compression as they provide superior image quality at high compression rates and yields lossy to lossless compression [3]. The lossy compression using wavelets far exceeds the lossy standard particularly at higher compression rates [4]. In compression, undesirable blocking artifacts are produced that are not present in wavelet compression. Lossless compression is realizable with the development of integer wavelets [5]. They generate wavelet coefficients that are integers unlike the conventional

2 wavelets, where the wavelet coefficients have to be truncated to the nearest integer resulting in the loss of floating point values. Various wavelet based coding methods have been developed to enhance compression, which includes development of coders that utilizes the spatial similarities among the wavelet coefficients. Such coders are called zero-tree coders, namely embedded zero tree wavelet coder (EZW) [6], and [7]. is an advancement of EZW and is the most popular coder. The prime advantage of using wavelets is that it supports progressive transmission capability useful for telemedicine [8]. In the case of traditional compression methods for images, it is required that the image has to be compressed to the desired quantity before transmission. Such techniques are unsuitable for telemedicine application as the amount of compression required cannot be known, and the file has to be recompressed at varying compression rates and retransmitted until the demand of the receiver is satisfied. On the other hand, the techniques that adopt progressive transmission, the data will be transmitted in successive stages until the required image quality is reached, after which the receiver can cease the process. At the receiving end, the vital features in the image can be realized quickly and with minimum bandwidth utilization. If all the information is transmitted, the reconstructed image will be identical to the original image. In most cases, it is not necessary to send the entire information because the diagnostic relevant features in the image are preserved even with lossy compression using wavelets. COMPRESSION USING is a progressive transmission coder that produces embedded bit-streams. It works on the principle of spatial relationship among the wavelet coefficients at different levels and frequency sub-bands in the pyramid structure of wavelet decomposition. This pyramid structure is commonly known as spatial orientation tree. If a given coefficient at location is significant in magnitude then some of its descendants will also probably be significant in magnitude. The algorithm takes advantage of the spatial similarity present in the wavelet space to optimally find the location of the wavelet coefficients that are significant by means of a binary search algorithm. Fig.1 shows the process of compression. The Transfer Syntax Unique Identification field of the header is modified to indicate that the image is compressed using. 2-D wavelet decomposition is applied to the image data. Biorthogonal (9-7) integer wavelet obtained from the lifting scheme presented in [5] is used for wavelet decomposition. The decomposed image is then coded using. produces embedded bit-streams with the output being a string of 0 s and 1 s. The header is finally added to the embedded bitstream resulting in the compressed image. Data Header Fig. 1. Compression of During transmission, the header and the image data are first separated from the compressed image. The header is first transmitted followed by the bit-stream. Based on the Transfer Syntax Unique Identification field of the header, appropriate compression method is identified. As the bit stream is being received, it is decoded by the decoder, and wavelet reconstruction is applied to obtain the reconstructed image. The receiver can terminate this process at any instant. The image quality at the receiving end depends on the number of bits received and decoded by. The reconstruction process is depicted in Fig. 2. Compressed 2-D LWT TSUID Modification Bit- Stream Header Encoder Decoder Fig. 2. Reconstruction of EXPERIMENTAL RESULTS Bit- Stream 2-D ILWT Compressed Data Experiments were carried out using two images, namely, X-ray chest image and MRI brain image (512 x 512) with 8-bit resolution.

3 Fig. 3 and 4 display these images reconstructed during the progressive transmission for varying bit rates. The performances are evaluated in terms of PSNR and. Though PSNR is a widely used evaluation parameter for image quality, it doesn t exactly indicate the perceived visual quality of the image. Hence, evaluation has also done using, [9], which is based very much on the characteristics of human visual system (HVS) and measures the structural similarity between two images. The structural similarity index is equal to one if the two images are identical. (i) (ii) For comparison purposes with our method, images were compressed and transmitted using the lossy compression schemes currently supported by the standard. Tables (1) and (2) show the PSNR of chest image and brain image for different compression schemes. Similarly, Tables (3) and (4) show the index of these two images for s. Table (1). PSNR of Chest image for Tabl e (2). PSNR of brain image for (iii) (iv) Fig. 3. (i) Original Chest image, (ii) (iv) Chest image reconstructed at 0.1, 0.5 and 1.0 bpp respectively (i) (iii) (ii) (iv) Table (3). of DICO M chest image for Fig. 4. (i) Original brain image, (ii) (iv) MRI brain image reconstructed at 0.1, 0.5 and 1.0 bpp respectively

4 Table (4). of brain image for Fig. 5 and 6 give the plot of bit rate against the PSNR and index respectively for various compression schemes for the chest image Bit Rate Fig. 5. Bit Rate versus PSNR for chest image bits received increases. From Table (1) and (2), it is found that PSNR value increases as the compression rate decreases. perform worst when subjected to high compression, but gives better results at low compression rate. Our method gives the highest PSNR value for all compression rates and performs better than the other compression standards. From Table (3) and (4), it is found that the index approaches one as the compression rate decreases. and performs consistently for all compression rates compared to and. This is because of the progressive transmission properties of and. Further it can be noted that the PSNR value does not correlate with the index for certain results. CONCLUSIONS In this paper, we have highlighted the advantages of progressive transmission for telemedicine applications by compressing images using wavelets and subsequently coding the coefficients using. The results indicate that in progressive transmission, the images are reconstructed from low resolution to high resolution enabling the receiver to view the entire approximation image with less information being transmitted, thereby, consuming less bandwidth and saving time. gives better results than and in general, progressive transmission methods of and perform well at higher compression rates compared to the other compression schemes Bit Rate REFERENCES 1. Standard, Fig. 6. Bit Rate versus index for chest image DISCUSSION From Fig. 3 and 4, it is evident that in progressive transmission the images are reconstructed from coarse resolution to finer resolution with each stage of transmission. The quality of the image increases as the number of 2. Part 5: Data structures and Encoding, 3. A.Said and W.A. Pearlman, IEEE Trans. on Processing, 5 (9), 1303, (1996). 4. T.A. Iyriboz, M.J. Zukoski, K.D. Hopper and P.L. Stagg, J Digit Imaging, 12, 14, (1999).

5 5. W.Sweldens, SIAM J. Math. Anal., 29, 511, (1998). 6. J.M.Shapiro, IEEE Trans. on Signal Processing, 41 (12), 3445, (1993). 7. A.Said and W.A. Pearlman, IEEE Trans. on Circuits and Systems for Video Technology, 6 (3), 243, (1996). 8. R. S. Dilmaghani, A. Ahmadian, M. Ghavami,and A. H. Aghvami, IEEE Signal Processing Letters, 11 (10), 806, (2004). 9. Zhou Wang, Alan Conrad Bovik, Hamid Rahim Sheikh, and Eero P. Simoncelli, IEEE Trans. on Processing, 13 (4), 600, (2004).