PERFORMANCE AND ANALYSIS OF WAVELET BASED MEDICAL IMAGE COMPRESSION USING EZW

8/10/2019 PERFORMANCE AND ANALYSIS OF WAVELET BASED MEDICAL IMAGE COMPRESSION USING EZW

1/6


2/6

[ 76 ]

dimensional imge there exist any compressiontechniques such as JPEG, BMP, GIF and thenew wavelet based PEG2000 standard. Allschemes above are used for two- dimensionalimages. In medical image compression, diagnosisis effective only when compression techniques

preserve all the relevant information neededwithout any appreciable loss of information, Incase with lossless compression. While lossycompression techniques are more efficient interms of storage and transmission needs because

of high compression ratio and the quality4. Inlossy compression, image characteristics areusually preserved in the coefficients of thedomain space in which the original image is

transformed5. The quality of the image aftercompression is very important and it must bewith in the tolerable limits which vary from

image to image and the method to method, hencethe compression becomes more interesting asa part of qualitative analysis of different typesof medical image compression techniques6.There are mainly two categories of compre-ssion namely: 1-Lossless or irreversible com-pression and 2-Lossy or reversible compression.Wavelet Transform (WT) represents an imageas a sum of wavelet functions with differentlocations and scales. Basis for wavelettransform can be composed of any function

that satisfies requirements of multiresolutionanalysis, it means that there exits a largeselection of wavelet families depending on thechoice of wavelet function. Among the mostpopular wavelets are Haar and Daubechies4.

A popular method of image compression,namely, the embedded zero tree wavelet. Itwas introduced in the ground breaking paperof Shapiro1. The EZW encoder is based onprogressive encoding to compress an image

into a bit stream with increasing accuracy. Thismeans that when more bits are added to thestream, the decoded image will contain moredetail.

Wavelet Transform :

WT represents image as a sum ofwavelets on different resolution levels. Awavelet is a compact function, i.e., outside acertain interval it vanishes. Implementationsare based on the fast wavelet transform, wherea given wavelet is shifted and dilated so as toprovide a base in the function space. In otherwords, a one-dimensional function is trans-formed into a two dimensional space, where itis approximated by coefficients that depend ontime and on scale. The WT is especially well

suited to analyze local variations such as thosein still images. Mallat2had made an importantcontribution to the application of wavelet theoryto multimedia by introducing multiresolution: thetransition from mathematical theory to filters.Multiresolution analysis is implemented viahigh-pass filters and low-pass filters. In thiscontext, the wavelet transform of a signal orimage can be realized by means of a filter bankvia successive application of a 2-channel filterbank consisting of high-pass and low-passfilters.

Medical Image Content

According to Sonja Grgic3, the imagecontent of quality irrespective to technicalparameters of the system. Due to this reason,the content of our test medical images shouldbe understood first. The modalities of medicalimages have been identified, which includesMRI, CT, X-ray and ultrasound images. Fromour observation, we realized that different


3/6

[ 77 ]

modalities of medical image have differentcharacteristics, i.e., noise that present in theimage, texture, intensity profile, sharpness, etc.These differences are very important forcompression and diagnosis purpose.

Choice On Wavelets

Important properties of wavelet functionsin image compression applications are compactsupport. In this paper, we consider, Haar waveletDaubecbies 4 wavelet. A Haar wavelet is thesimplest type of wavelet. One distinctivefeature that the Haar transform enjoys is thatit lends itself easily to simple hand calculations.For the Daubechies 4 wavelet transforms, thescaling functions and wavelets have slightlylonger supports, i.e., they produce averages

and differences using just a few more valuesfrom the signal.

2. Embedded zero tree wavelet coding

EZW encoder is based on progressiveencoding to compress an image into a bit streamwith increasing accuracy. When more bits areadded to the stream, the decoded imagecontains more details of the image. Coding animage using the EZW scheme, together withsome optimizations, leads to results in aremarkably effective image compressor withthe property that the compressed data streamcan have any bit rate desired. Any bit rate isonly possible if there is information losssomewhere so that the compressor is lossy.However, lossless compression is also possiblewith an EZW encoder, with less optimal results.The design unit implements the EZW codingsystem for data compression. The codingsystem reads the multiresolution component ofthe image obtain from the transformation

module and pass the data to the decoder unitto retrieve the image back. Figure 6 shows theimplemented EZW coding system for imageprocessing.

Before the processing of image datathe image are preprocessed to improve the rateof operation for the coding system. Underpreprocessing tiling on the original image is

carried out. The term "tiling" refers to thepartition of the original image into rectangularnon overlapping blocks, which are compressedindependently, as though they were entirelydistinct images. All operations, includingcomponent mixing, wavelet transform,quantization and entropy coding are performedindependently on the image tiles. Tiling reducesmemory requirements, and since they are alsoreconstructed independently, they can be usedfor decoding specific parts of the image insteadof the whole image. All tiles have exactly thesame dimensions, except maybe those at theboundary of the image. Arbitrary tile sizes areallowed, up to and including the entire image.This unit transforms the input image from timedomain to frequency domain and decomposesthe original image into its fundamentalcomponents.

The wavelet transform uses filterbanks for decomposition of preprocessedoriginal image into 3 details and 1 approximate


4/6

coefficient. The filtering is carried out byconvolving the input image with the filtercoefficients passed. EZW encoder encodesthe decomposed image by recognizing thepriority of decomposed image pixel. Theencoder module calculates a initial threshold

for coding given by T0= 2^(log2xmax). Theencoding process is performed using 2 passesnamely dominant pass and subordinate pass.The dominant pass scans the coefficient usingthe threshold and assigned each coefficient witha symbol. Basically there fore isolated symbolsfor coding, they are positive significant (PS),negative significant (NS), isolated zero (IZ) andzerotree root (ZR). The other pass made atthe encoding unit is the subordinate pass wherethe coefficients are encoded as 0 or 1 dependingon the current threshold. These passes are

repeated for n cycles reducing the currentthreshold by 2 until the required data bit rate isreached.

The decoding unit reconstructs thevalues by identifying the symbols as positive,negative, zerotree and isolated zerotree. Inversetransformation is the process of retrieving backthe image data from the obtained image values.The image data transformed and decomposedunder encoding side is rearranged from higherlevel decomposition to lower level with thehighest decomposed levels arranged at the top.

a) The Algorithm

The EZW output stream starts withinformation to synchronize the decoder. Theminimum information required by the decoderfor its functionality is the number of wavelettransform levels used and the initial threshold.Basically, a constant level (3) of wavelettransform is used for transformation. The first

step in the EZW coding algorithm is todetermine the initial threshold. The initialthreshold t0 is given as t0=2 [log2(MAX(|g(x,y)|))].

where MAX(|(x,y)|) means the maximum

coefficient value in the image and (x,y)denotes the coefficient. Then taking the

obtained threshold as the initial value the scaledsub-band samples are subjected to dominantpass and subordinate pass. Under each pass,the threshold is decreased by half the value.This comparison is carried out until thethreshold reaches to the minimum threshold.The flow chart of the algorithm is presentedin figure 1.

Figure-1 flow chart EZW Algorithm

A wavelet transform represents asignal from the time domain in to the joint time-scale domain. i.e. the wavelet coefficients aretwo-dimensional. To compress the transformedsignal, not only the coefficient values but alsotheir position in time has to be coded. Whenthe signal is an image, then the position in timeis better expressed as the position in space.After wavelet transforming an image it can be

[ 78 ]


5/6

represented using trees because of the subsampling that is performed in the transform. Acoefficient in a lower subband can be thoughtof as having four descendants in the nexthigher subband as shown figure 4. The fourdescendants each also have four descendantsin the next higher subband, which gives a quad-tree, with every root having four leafs. Azerotree is defined as a quad-tree of which allnodes are equal to or smaller than the root andthe root is smaller than the threshold againstwhich the wavelet coefficients are currentlybeing measured. The tree is coded with a singlesymbol and reconstructed by the decoder as aquad-tree filled with zeroes. The EZW encodercodes the zero-tree based on the observationthat wavelet coefficients decrease with scale.In a zerotree all the coefficients in a quad treeis be smaller than the threshold if the root is

smaller than this threshold. Under this case thewhole tree can be coded with a single zerotree(T) symbol.

Figure-4 The inter relationship of multiplelevels in wavelet scaled image

EZW encoding uses a predefined scan order toencode the position of the wavelet coefficients.Through the use of zerotree many positionsare encoded implicitly. Several scan orders arepossible, as long as the lower subbands arecompletely scanned before going on to thehigher subbands. The relations betweenwavelet coefficients in different subbands, andthere scan path is show in figure 5.

Figure-5 Scanned order for embeddedcoding Results

RESULTS

Original Image

Harr DB4 Scaled Image Scaled Image

Final Image of Harr and DB4

Process Time %Error

[ 79 ]


6/6

3. CONCLUSIONS

In this paper, we found that wavelet-based medical image compression preferssmooth functions of relatively short length. Ourresults show that different wavelet filters

performed differently for different medicalimages, but generally the difference betweeneach other was not great. As the results shown,db4 is the best-suited wavelet filter for MRIimage in all compression bit rate. Meanwhile.From our observation, we found that Haar waveletperformed the worst for MRI and X-rayimages , but it shows a competitive result forUltrasound image. As conclusion, the choiceof the best performing wavelet filter in medicalimage compression is mostly depends on theimage content.

REFERENCES

1. J. Shapiro, "Embedded image coding usingzerotree of wavelet coefficients "IEEETrancs. Sighnal processing Vol. 41, pp.3445-3462, Dec (1993).

2. Stephane Mallat, "A Compact Multiresolution

representation: The wavelet model, "IEEEComputer Society Workshop on computervision (WCV), 87, 2-7 (1987).

3. Sonja Grgic, Mis lav Grgic and BrankaZovko-Cihlar, "Performance Analysis ofImage Compression Using Wavelets",

IEEE Trans. On industrial Elecfronics, vo1.48, No. 3, June (2001).

4. Guest Editorial, "Wavelets in Medical Imaging",IEEE Trans on Medical Imaging, Vol. 22,No. 3, March (2003).

5. Yung-Gi Wu, Shen-Chuan Tai, "MedicalImage Compression by Discrete CosineTransform Spectral Similarity Strategy",IEEE Transactions on Information Technology

in Biomedicine, Vol. 5, No. 3, September(2001).

6. S. Tai, Y. Wu and C. Lin, "An adaptive 3-DDiscrete Cosine Transform Coder for medicalimage compression," IEEE Trans. Inform.Technol. Biomed., Vol. 4, pp. 259-263, Sept.(2000).

[ 80 ]

Documents

PERFORMANCE AND ANALYSIS OF WAVELET BASED MEDICAL IMAGE COMPRESSION USING EZW