Compression of Magnetic Resonance Imaging and Color Images Based on Decision Tree Encoding
Total Page:16
File Type:pdf, Size:1020Kb
International Journal of Pure and Applied Mathematics Volume 118 No. 24 2018 ISSN: 1314-3395 (on-line version) url: http://www.acadpubl.eu/hub/ Special Issue http://www.acadpubl.eu/hub/ Compression of magnetic resonance imaging and color images based on Decision tree encoding D. J. Ashpin Pabi,Dr. V. Arun Department of Computer Science and Engineering Madanapalle Institute of Technology and Science Madanapalle, India May 24, 2018 Abstract Digital Images in the digital system are needed to com- press for an efficient storage with less memory space, archival purposes and also for the communication of information. Many compression techniques were developed in order to yield good reconstructed images after compression. An im- age compression algorithm for magnetic resonance imaging and color images based on the proposed decision tree encod- ing is created and implemented in this paper. Initially the given input image is divided into blocks and then each block are transformed using the discrete cosine transform (DCT) function. The thresholding quantization is applied to the generated DCT basis functions, reduces the redundancy on the image. The high frequency coefficients are then encoded using the proposed decisiontree encoding which is based on the run length encoding and the standard Huffman cod- ing. The implementation using such encoding techniques improves the performance of standard Huffman encoding in terms of assigning a lesser number of bits, which in turn reduces pixel length. The proposed decision tree encoding also increases the quality of the image at the decoder. Two data sets were used for testing the proposed decision tree 1 International Journal of Pure and Applied Mathematics Special Issue encoding scheme. A first set of experiments was made on the medical MRI images and later to the color images. The accuracy of the proposed method depends on the factors block sizes, threshold values and the final decision tree en- coding. Experimental result shows the effectiveness of the proposed method over the other existing methods. Key Words:image compression; discrete cosine trans- form; magnetic resonance imaging; color images;quantization; run-length encoding;decision tree encoding; Huffman encod- ing. 1 Introduction Medical and color images need higher number of bits per color channel than the traditional images. That leads the problem in storage and transmission. Thus, compression is appliedto reduce the amount of bits required to store an image. Images in the digi- tal systems are represented as pixels. Combination of pixels forms an image. Each pixel occupies a particular location in an image. The intensity of those pixels was variable. Each pixel in an im- age is represented by the number of bits. Images consist of pixels that are highly correlated, that occupies massive storage space and minimizes the transmission bandwidth. The challenge in image compression is to represent those pixels with a minimum number of bits. The primary role of image compression is the reduction of the amount of image data (bits) while preserving image details. Three types of data redundancy are observed [1]. Image compression techniques are broadly categorized into two classes: statistically redundant (lossless) or visually irrelevant (lossy) compression techniques. Removing the visually irrelevant informa- tion result losses on the reconstructed output image. Thus lossy image compression generates reconstructed image with some data loss, whereas lossless keeps image information. The lossy compres- sion techniques achieve a high compression ratio. The removal of statistically redundant data is called as reversible or lossless com- pression techniques. In which the image data can be completely retrieved from the compressed data. Thecompressed reconstructed image of the lossless imagecompressor corresponds to the original image. Medical imaging produces a digital picture of human body 2 International Journal of Pure and Applied Mathematics Special Issue [2]. They are diagnosed by a radiologist before they are made re- viewed by other members of the medical staffs. At the time of diagnosis the radiologists interpret the images and dictate the di- agnostic report. Large medical institution produces 5, 10 or even 100GB of radiological image data each day. The proposed com- pression algorithm is passed to the medical image data set and their performance is evaluated by comparing the statistical values of other compression methods. In this digital era, the need of color images has grown high. In internetmore than 90 percent of the data isalso comprised of color images [3]. Thus the need of color image compression increases. Each color component in a digital RGB color image requires 8 bit data. These color components contain redundant data, which re- quires large amount of storage space. To store color images with minimum number of bits those images has to be compressed. Thus, the second data set consists of color JPEG images are also used in this paper. An RGB color image requires three bytes for each pixel. The challenge in color image compression is to reduce the correla- tion between the R, G and B planes.The rest of the paper contains five sections: Section 2 describes related works and contribution. Section 3 describes the proposed compression technique. Section 4 presents experimental results. Section 5 draws the conclusion of the proposed work and the possible works. 2 related work and contribution There is a great concern between the research community in the implementation of image compression schemes over Magnetic res- onance imaging and color images. The admissible parts of these researchers aim to minimize storage space with acceptable quality. A BEND based image compression method introduced in [4] in- corporates the dowmsampling technique. A cascade feedback com- pression framework is constructed based on the spatial similarity property of the BIMFs at different resolutions. In his work, the spatial redundancy is removed by estimating BIMFs at high reso- lution levels. Reference [5] developed a real time solution to provide flexibility of the resulting image while transmit the image content to web pages. The performance of his real-time system depends 3 International Journal of Pure and Applied Mathematics Special Issue on the adjustable parameters and verified in terms of compression ratio, processing delay and the quality of the compressed image. An adaptive spatial post processing algorithm is introduced and it is applied to the block based discrete cosine transform (BDCT) coded images [6]. Sparsifying transform such as wavelets and DCT is used to compress images [7]. Theyhave done four different initial- izations. The first is the application of 6464 2D DCT matrix. The second is done by inverting/transposing the left singular matrix. The third and the fourth initializations were the identity matrix. Reference [8] invented a compression method and which is consid- ered as the extension of the warped discrete cosine transform known as the 3-D warped discrete cosine transform (3-D WDCT). Refer- ence [9] designed a technique for medical image compression that improves the performance of JPEG 2000 for a volumetric set of medical images. The possibility of using the particle swarm optimization (PSO) technique in [10]. A set of optimal thresholds was obtained using the PSO algorithm. At last the variable length encoding scheme was used to encode the results. Such a method of selecting an op- timal threshold value based on the PSO algorithm resulted with less entropy leads to a gain in compression. An improved medical image compression based on region of interest (ROI) is developed in [11]. The image was split into two parts: ROI regions and non- ROI regions. A Lossless compression algorithm was applied to ROI regions and lossy to rest of other regions reduces the complexity in compression. Reference [12] proposed a lossy image compression algorithm based on the DCT transform and an adaptive block scan- ning. The efficiency of their scheme was demonstrated by results over the color images.A new lossless chain code compression based on move-to-front transform and an adaptive run-length encoding is designed in [13]. Their proposed adaptive run-length encoding optimized the number of bits. 3 the proposed decision tree encoding compression technique The flow of the proposed decision tree based image compression/decompression model is shown in figure 1. Figure 1 (a) depicts the different 4 International Journal of Pure and Applied Mathematics Special Issue stages of the proposed decision tree based compression technique, (b) shows the stages of the decompression phase. The proposed compression algorithm consists of the following 3 stages. Block based DCT Transformation Thresholding quantization Implementation of the proposed decision tree encoding The input to the system is a digital magnetic resonance imag- ing or color images and the output is the compressed one. First an image is given as an input and then it was partitioned into 88 or 1616 or 3232 non-correlated blocks and the DCT transform [14] was applied. The DCT coefficients were rounded using the thresholding quantization [15] and then the proposed decision tree based encoding was performed. The zig-zag scanning is done while encoding the DCT coefficients. The compressed bit streams were the output from the compressor. The reconstructed image was ob- tained fromthe decompressor and the steps for the decompression are shown in Fig.1 (b). A Block diagram of the proposed decision tree based compression