Comparison of Image Compression and Enhancement …1150997/FULLTEXT02.pdf · data loss is used as the best method for image compression of medical images. Lossless ompression ﬁnds

Master ThesisElectrical EngineeringFebruary 2017

Master of Science in Electrical Engineering withEmphasis on Signal Processing

Comparison of Image Compressionand Enhancement Techniques forImage Quality in Medical Images.

Submitted by

Sai Virali Tummala

Veerendra Marni

Department of Applied Signal ProcessingBlekinge Institute of TechnologySE–371 79 Karlskrona, Sweden

This thesis is submitted to the Department of Applied Signal Processing at Blekinge Instituteof Technology in partial fulfillment of the requirements for the degree of Master of Science inApplied Signal Processing.

Contact Information:

Author(s):

Sai Virali TummalaE-mail: [email protected]

Veerendra MarniE-mail: [email protected]

University adviser:

Irina GertsovichDepartment of Applied Signal Processing

Dept. Applied Signal Processing Internet : www.bth.seBlekinge Institute of Technology Phone : +46 455 38 50 00SE–371 79 Karlskrona, Sweden Fax : +46 455 38 50 57

Abstract

Context: Image Processing is the processing of images, series of im-ages or videos by using mathematical operations by using any form ofsignal processing techniques. Image Compression and Image Enhance-ment are the most widely used techniques in Image Processing. Now aday there is an increasing need of these techniques in the medical field.This thesis is focused on the performance quality comparison of med-ical images using Image Compression and Enhancement Techniques.This analysis is used to suggest the better techniques for compressionand enhancement of medical images.Objectives: In this study, the main objective is to attain an efficientoutput of a medical image. This undergoes a series of steps startingwith compression and then followed by the enhancement of the medi-cal image to get an enhanced output. We provide a detailed analysisof all the techniques involved in this process. The images quality isthen assessed on various performance parameters.Methods: A detailed literature research has been done to study thevarious techniques in both image compression and enhancement. Theperformance metrics are considered by understanding the literatureresearch from various papers. Both the lossy and lossless methods areused in image compression. The lossy technique has been done in bothDiscrete Cosine Transform (DCT) and Discrete Wavelet Transform(DWT) and lossless in both Run Length Encoding (RLE) and BlockTruncation process. Further, the enhancement of the compressed out-puts is performed using Image Intensity Adjustment, Adaptive His-togram Equalization (AHE) and Morphological Operations. The re-sults are obtained and the performance metrics are compared. MAT-LAB is used for the coding purpose.Results: The results are calculated using the performance metricsPSNR, MSE and SSIM where the values for each technique are tab-ulated. Then plots for each image and for each performance metricare plotted. Through these outputs we compare different performanceparameters by adjusting the coefficients and also the block sizes.Conclusions: With a detailed analysis and logical comparison of theperformance metrics we conclude the better performance metric thanthe other and also which combinations of compression and enhance-ment techniques are better with each other.Keywords: AHE, Block Truncation Process, DCT, DWT, ImageCompression, Image Enhancement, Morphological Operations, RLE.

i

Acknowledgement

We would like to express our deep sense of sincere gratitude to oursupervisor Irina Gertsovich for her continuous support and guidancein each and every aspect to achieve the aim of this thesis work. Thishelped us enhance our skills and the quality of our thesis. I would alsolike to extend my gratitude towards my University, BTH, because ofwhich we are what we are today.

Also, we would like to heart fully thank our friend Suveen KumarVundavalli and Sri Krishna Jayanthi who extended their support di-rectly or indirectly for us in building up this project. And also we liketo thank all our friends for their continuous love and support.

And also finally, last but not the least, we would like to thank ourparents for making us what we are today. Their constant support,love and encouragement have made us achieve all our goals in our lifein a better way.

Thank You!!!

ii

Abbreviations

AHE Adaptive Histogram Equalization

BTC Block Truncation Coding

CR Compression Ratio

DCT Discrete Cosine Transform

DWT Discrete Wavelet Transform

HL High low

HH Low High

JPEG Joint Photographic Experts Group

LL Low Low

LH Low High

MO Morphological Operations

MRI Magnetic Resonance Imaging

MSE Mean Square Error

PSNR Peak Signal to Noise Ratio

RLE Run Length Encoding

SSIM Structural Similarity Index Modulation

iii

Contents

Abstract i

1 Introduction 11.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2 Aims and Objectives . . . . . . . . . . . . . . . . . . . . . . . . . 21.3 Research Questions . . . . . . . . . . . . . . . . . . . . . . . . . . 21.4 Documentation Framework . . . . . . . . . . . . . . . . . . . . . . 3

2 Related Work 4

3 Methodology 63.1 Theoreotical Background . . . . . . . . . . . . . . . . . . . . . . . 63.2 Image Compression Techniques . . . . . . . . . . . . . . . . . . . 6

3.2.1 Lossy Techniques . . . . . . . . . . . . . . . . . . . . . . . 83.2.1.1 Discrete Cosine Transform (DCT) . . . . . . . . 8

3.2.2 Discrete Wavelet Transform (DWT) . . . . . . . . . . . . . 93.2.3 Lossless Compression . . . . . . . . . . . . . . . . . . . . . 9

3.2.3.1 Run Length Encoding (RLE) . . . . . . . . . . . 103.2.3.2 Block Truncation Coding . . . . . . . . . . . . . 10

3.2.4 Enhancement Techniques . . . . . . . . . . . . . . . . . . . 113.2.4.1 Adaptive Histogram Equalization (AHE) . . . . . 11

3.2.5 Morphological Operations (MO) . . . . . . . . . . . . . . . 123.3 Performance Metrics . . . . . . . . . . . . . . . . . . . . . . . . . 12

3.3.1 Peak Signal to Noise Ratio (PSNR) . . . . . . . . . . . . . 123.3.2 Mean Square Error (MSE) . . . . . . . . . . . . . . . . . . 133.3.3 Structural Similarity Index Modulation (SSIM) . . . . . . 13

4 Results & Analysis 144.1 Output Image for 111.tif . . . . . . . . . . . . . . . . . . . . . . . 144.2 Output Image for 222.tif . . . . . . . . . . . . . . . . . . . . . . . 174.3 Output Image for 333.tif . . . . . . . . . . . . . . . . . . . . . . . 204.4 Tabular Forms of Performance Parameters . . . . . . . . . . . . . 24

4.4.1 Performance Metrics Tabular Form for 111.tif . . . . . . . 244.4.2 Performance Metrics Tabular Form for 222.tif . . . . . . . 25

iv

4.4.3 Performance Metrics Tabular Form for 333.tif . . . . . . . 264.5 Plots for PSNR, MSE and SSIM for the image database . . . . . . 27

5 Conclusion & Future Work 31

References 32

v

List of Figures

3.1.1 Block Diagram of the Compression and Enhancement Process. . . 7

4.1.1 Lossy compression of medical image using DCT . . . . . . . . . . 144.1.2 Lossy compression of medical image using DWT . . . . . . . . . . 154.1.3 Enhancement of DCT compressed image using AHE and MO . . . 154.1.4 Enhancement of DWT compressed image using AHE and MO . . 154.1.5 Lossless Compression using BTC and RLE . . . . . . . . . . . . . 164.1.6 Enhancement of BTC compressed image using AHE and MO . . . 164.1.7 Enhancement of RLE compressed image using AHE and MO . . . 174.2.1 Lossy compression of medical image using DCT . . . . . . . . . . 174.2.2 Lossy compression of medical image using DWT . . . . . . . . . . 184.2.3 Enhancement of DCT compressed image using AHE and MO . . . 184.2.4 Enhancement of DWT compressed image using AHE and MO . . 184.2.5 Lossless Compression using BTC and RLE . . . . . . . . . . . . . 194.2.6 Enhancement of BTC compressed image using AHE and MO . . . 194.2.7 Enhancement of RLE compressed image using AHE and MO . . . 204.3.1 Lossy compression of medical image using DCT . . . . . . . . . . 204.3.2 Lossy compression of medical image using DWT . . . . . . . . . . 214.3.3 Enhancement of DCT compressed image using AHE and MO . . . 214.3.4 Enhancement of DWT compressed image using AHE and MO . . 214.3.5 Lossless Compression using BTC and RLE . . . . . . . . . . . . . 224.3.6 Enhancement of BTC compressed image using AHE and MO . . . 224.3.7 Enhancement of RLE compressed image using AHE and MO . . . 234.5.1 PSNR plot for all medical images database. . . . . . . . . . . . . 284.5.2 SSIM plot for all medical images database. . . . . . . . . . . . . . 29

vi

List of Tables

4.1 Image 111.tif tabular form . . . . . . . . . . . . . . . . . . . . . . 244.2 Image 222.tif tabular form . . . . . . . . . . . . . . . . . . . . . . 254.3 Image 333.tif tabular form . . . . . . . . . . . . . . . . . . . . . . 26

vii

Chapter 1Introduction

As in today’s world of emerging technology where most of the data is recordedin digital format, virtually all image interpretation and analysis involves someelements of digital processing. This digital image processing involves the pro-cessing of images, series of images or videos by using mathematical operationsby using any form of signal processing techniques. Image compression and imageenhancement are the most widely used techniques in image processing. Differenttypes of images like binary images, indexed or pseudo colored images, grayscaleimages, true color images also known as RGB images are generally used in imageprocessing. Image Processing is of generally digital image processing but thereare also analog and optical image processing possible.image compression is anefficient technique to reduce the size of graphical file and also reduce the storagerequirement area [1].

Medical images like Magnetic Resonance Imaging (MRI) scans, X-ray imagesare the most used images these days in the medical field. As there is an emerginggrowth of population these days so are the health issues of the people. The differ-ent cases of number of patients and their records are maintained in the hospitals.So, for storing all the case history of a patient there are a number of medicalimages that has to be stored in the system database. In this regard the medi-cal images are compressed using several techniques and thus images are storedand transmitted from one system to another for the ease of communication. So,in this project the medical images undergo both compression and enhancementtechniques consecutively. And later on the performance quality of the images aretested on different performance parameters which are mostly used in to check theperformance of compression and enhancement techniques.

1.1 MotivationImage compression and image enhancement techniques are the most widely usedtechniques these days in the field of medical images. So, in this master thesis wecompare the performance quality of the different compression and enhancement

1

Chapter 1. Introduction 2

techniques based on different performance metrics. The basic idea is to considerdifferent medical images and perform compression techniques on the images. Thenthe images are again restored back by enhancing them. Then, we calculate theperformance measuresniques of the compression and enahancement tech by usingdifferent performance parameters like peak signal to noise ratio (PSNR), meansquare error(MSE) and structural similarity index modulation(SSIM).

Our main motive of this thesis is to compare different techniques on the samemedical images and see how the performance varies on different combinationsaccordingly.thus the quality of the output image is compared with the inputimage and the performance of the combinatons are analysed.

1.2 Aims and ObjectivesThe main aim of this thesis is the performance quality comparison of medical im-ages using both image compression and enhancement techniques. The objectivesmainly include:

• Selection of necessary medical image database from the open source librariesavailable.

• Compression of the medical images using both lossy and lossless compressiontechniques.

• Enhancing the compressed images using different enhancement techniques.

• Comparison and evaluation of the quality of the obtained output images ineach case with respect to the original image.

1.3 Research QuestionsThe research questions discussed in this thesis are :

• What are the methods or techniques used for image compression?

• What are the methods or techniques used for image enhancement?

• What are the performance metrics that need to be considred to comparethe performance results of different combinations of image compression andenhancement methods?

Chapter 1. Introduction 3

1.4 Documentation FrameworkThe document is organized as mentioned below.

Chapter 1 gives a brief introduction about the thesis in which way imageprocessing is being used widely these days in the field of medical images. Thissection also deals with the main motive of the thesis and the aims and objectivesof this thesis. This is further followed by research questions and documentationframework.

Chapter 2 discusses the various projects or papers that are already publishedon the image compression and image enhancement of medical images.

Chapter 3 mainly focuses on the various methods or techniques used for theperformance quality comparison of medical images. Here we first discuss aboutthe literature review that has been done for this thesis project and then followedby compression and enhancement techniques in this project in detail. Then theperformance metrics are analyzed one after the other and how each metric is im-plemented.

Chapter 4 includes the results and the analysis part of each technique. Thevalues that are obtained from each performance metric are tabulated and therespective graphs plotted for each metric are included in this section. The resultsare analyzed and validated.

Chapter 5 gives a clear conclusion of the thesis project based on the aboveanalyzed and validated results. The future scope of this thesis is also mentionedhere in this section.

Chapter 2Related Work

This thesis mainly focuses on the concepts of image compression and image en-hancement techniques.In this thesis project work a literature review has beenmade in order to assess the progress made in the field of image compression andimage enhancement techniques on medical images. In [2], the author mainlyconcentrates on the types on compression techniques available for medical imagecompression and their classifications. In this paper the author also used perfor-mance parameters for the comparison of the images compressed. These techniqueshave been used for comparing different compression techniques and the values ofpeak signal to noise ratio (PSNR) are calculated.

In [3] , we will come to know a critical review on the image enhancementtechniques that are being used for the medical gray scale images. In this paper thetechniques are classified into frequency and spatial domains and the advantagesand disadvantages of these techniques are discussed.

In [4], the author focuses on developing some simple functions to computediscrete cosine transform (DCT) and compress the images. The 2D DCT is usedfor the compression of images in this paper. The author uses one dimensional andtwo dimensional DCT as well. Here, the image is converted into 8*8 block ma-trix for compression and the quantization technique is also used for compressionprocess. This entire process is followed by inverse 2D DCT for reverse decodingof the image that has been compressed.

In [5], the author uses 2D Discrete Wavelet Transforms to decompose theimage both spatial and spectral coefficients. Here, the image is divided into 4parts of low-low(LL), low-high(LH), high-low(HL), and high-high(HH). And theimages are compressed using discrete wavelet transform (DWT) and comparedusing performance parameters. In [6], the author compresses the images usingDWT and then compares the performance using metrics like PSNR, mean squareerror (MSE) and compression ratio (CR).

In [7], the author uses a hybrid combination of both DCT and DWT for thecompression of medical gray scale images. Here the author shows that DWTwith a two-threshold method named "improved-DWT" provides a better qualityof image compared to DCT and to DWT with a one-threshold method. Finally,the combination of the two techniques, named improved-DWT-DCT compres-

4

Chapter 2. Related Work 5

sion technique yields a better performance than DCT –based joint photographicexperts group (JPEG) in terms of PSNR.

In [8], the author shows a comparison between the RLE and Huffman algo-rithms for lossless data compression of medical images. This study points to theeffectiveness of the algorithm in the process of reducing the size of the files.

For the further study in the compression techniques, the author in [9], usedsome modified block truncation coding along with other algorithms to compressthe image. In this case this method has provided an image that is more robustand requires very little error protection overhead.

And for the enhancement of the compressed images the author in [10], hasproposed an image enhancement technique for image contrast enhancement usinga histogram modified framework and its applications. The experimental resultsshow the effectiveness of the algorithm in comparison to other enhancement al-gorithms.

The author in [11], also used morphological filtering for image enhancement forcleaning the image from various types of noise using the morphological operationslike erosion, dilation for the enhancement of the images.

Chapter 3Methodology

3.1 Theoreotical BackgroundThe performance metrics that are to be used for comparison of the compressedand enhanced images were selected by reading several related research papers andjournals. The methodology to reach the aim of this project involves an experi-mentation part from which the required data and graphical representations canbe obtained to make an appropriate analysis. The experimentation includes col-lecting the values of the selected performance metrics and plotting the graphs forthe values obtained. The experiments were performed using MATLAB software.

For this purpose we need medical images to test and compare the techniques.So a medical database of nearly one hundred images is collected from an opensource in the Internet.This database is a collection of 100 medical gray scale im-ages taken from different open sources for testing in the code. The dimensionsof the images taken are nearly .The images that are collected are free from thecopyrights issues and are open for the public to use them.

The block diagram in fig 3.1 depicts the research methodology followed in thisthesis to achieve the desired goal of this thesis.

3.2 Image Compression TechniquesImage Compression addresses the problem of reducing the irrelevance and redun-dancy of the image data in order to be able to store or transmit data in efficientform. As there is a wide growth in medicine field in day to day life there is alsoa great need for image compression techniques for storing abundant data andinformation. Image Compression is nothing but the size of the image is actuallyreduced in size without degrading the quality of the image[12]. The reduced filesize thus helps in storing more number of images in a file and for easy sendingand communication to others[13].

6

Chapter 3. Methodology 7

Figure 3.1.1: Block Diagram of the Compression and Enhancement Process.

There are several ways in which images can be compressed. Image can becompressed either with or without data loss. Depending on whether the data islost or not image compression is mainly of two types,

• Lossy Compression

• Lossless Compression

There are many techniques in both lossy and lossless techniques. But in ourthesis project we considered comparing only two lossy and two lossless techniquesrespectively.

The techniques used in lossy image compression are,

• Discrete Cosine Transform (DCT)

• Discrete Wavelet Transform (DWT)

The techniques used in lossless image compression are,

• Run Length Encoding (RLE)

• Block Truncation Coding (BTC)


3.2.1 Lossy Techniques

Lossy compression techniques are those techniques in which the compression ofthe image is done with the loss of some information. The compressed image lookssimilar to the original image but there is some loss in the information which can bedifficult to see in the compressed image. The lossy techniques that are used in thecompressing schemes such as jpg, png etc. In this thesis we considered several grayscale medical images and compressed them. Lossy compression comparatively hashigher compression ratio than the lossless techniques. Performance of the lossytechniques are mainly measured by such metrics as compression ratio, signal tonoise ratio and speed of encoding and decoding. The techniques used in thisproject for the lossy compression of medical images are,

• Discrete Cosine Transform (DCT)

• Discrete Wavelet Transform (DWT)

3.2.1.1 Discrete Cosine Transform (DCT)

The main objective of the image compression systems based on transform tech-niques is to store data efficiently and also to provide a good tradeoff between thecompression rates and the signal to noise ratios. In this thesis we have consideredDCT as DCT gives better results in terms of mean square error and compressionratio values compared to any other technique for gray scale medical images[4].DCT is in the base of JPEG image compression. DCT is also fast compared toothers and is also best for images with smooth edges. It transforms a signal fromits spatial domain to frequency domain. The images after reconstruction are in-versely proportional to the values of quantization. It packs the most importantinformation into few coefficients.

A gray scale medical image is taken and compressed using DCT and inverseDCT is used for reconstructing back. This process is done twice.This process ofcompression is done twice so as to reduce the spatial resolution of the image inthe first step and after this the image is divided into blocks and compressed againin the second step.So the first step is done using matlab formula,

And the secondly, the image is split into blocks of 8 by 8 where each blockundergoes 2dct.

The encoding and decoding process follows for full compression process usingIDCT to support 8 by 8 pixel per precision. In this compression all the coeffi-cients from the top left corner in the matrix are considered so we have taken anumber i.e., 20000 so that the high data is compressed well. So, after the generalcompression and decompression process the output may not be in the originalrange (0, 255), so the output is resized. So, thus the out compressed image isobtained which is compared with the original input for errors.


3.2.2 Discrete Wavelet Transform (DWT)

One of the most widely used transform techniques for image compression of med-ical images using wavelets is Discrete Wavelet Transform (DWT). This DWTis very useful for compressing signal and also shows better results for medicalgray scale images. While using DWT the important parameters that are takeninto consideration are testing the image, wavelet function, number of iterationsand calculation complexity. These wavelets transforms are used to process andimprove signals in fields like medical imaging where image degradation is not tol-erated.

The same input image which is taken earlier for DCT is now compressed usingthe DWT compression technique. The image is converted from mat to gray andthen it is divided into 4 bits in the form of (low, low), (low, high), (high, low),(high, high). The image is undergone through DWT compression and then theimage is again resized to original size. In this way the image is compressed us-ing DWT. The performance metrics are then calculated using PSNR, MSE, andSSIM and are tabulated for the further comparison with the other techniques.

3.2.3 Lossless Compression

The lossless compression technique is the other most important techniques inimage compression techniques. In this lossless compression technique, the com-pression of the image is done without incurring any major data loss in the image.This means that the image will be compressed but there will no significant loss indata which means all the useful information is not removed through compression.This lossless compression due its capacity of compressing the image without anydata loss is used as the best method for image compression of medical images.Lossless ompression finds its great use in medical field due to its rapid growthin recent times. As the number of hospitals and number of case records are in-creasing day by day there is also an increasing growth in need for compression ofimages for their storage and easy transmission. So, in these cases a case record isvery important to be stored in a compressed format and also without any loss indata because of compression. In this way lossless compression satisfies both thecases. Nevertheless, this only comes at the expense of obtaining low compressionrate values compared to lossy techniques. Most lossless compressions use entropyencoding methods for the compression.


The lossless compression that we used here in our thesis for compressing thegray scale medical images and to compare the performance metrics are,

• Run Length Encoding (RLE)

• Block Truncation Coding (BTC).

3.2.3.1 Run Length Encoding (RLE)

In the lossless compression techniques, one of the most widely used encodingtechniques is Run length encoding (RLE). RLE technique actually compressesthe medical images without losing the important information or data. This tech-nique compresses the images with a continuous long sequence into a single datasequence. Run Length Encoding is mostly in use din compressing black and whiteimages as this gives better results in compression of images.

In this thesis, the medical image is selected from the medical images databasefrom an open source and is tested using matlab. The image is first converted frommat to gray and is given as an input for compression. Image intensity adjustmentalgorithm is also used here to enhance the contrast of the image. This algorithmdoes not provide any significant change in the original file. The loaded imageis further converted into the desired form and the for loop is implemented. Theiteration is repeated as long as the coefficients of the images used are iterated andthe image is compressed into a single sequence. Then the image is iterated andthen the loop is removed. Then the RLE out compressed image is obtained andthe lossless compression using RLE is obtained.

3.2.3.2 Block Truncation Coding

In this thesis project we used Block Truncation Coding as another compressionmethod for gray scale medical images. This method is used as one of the losslesscompression technique for medical images. In many cases, RLE and BTC for loss-less compression are used together as a combination for achieving the compressionoutputs. This technique is implemented as a set of nodes and can be easily storedas a regular set of arrays [14].Here in this project we used block truncation codingas another compression method for gray scale medical images. This method isused as one of the lossless compression technique for medical images.

In this project the images from the open source medical image database is se-lected and taken as the input. The block size of the images is adjusted accordingto which we gt the desired output. We used BTC in some parts so as to allowthe separation of the image into blocks. This technique is followed by column


filtering so that the entire column values are adjusted.The BTC approach hasthe advantage of being extremely easy to implement; moreover, it often possessesgood performance characteristics relative to other techniques even in the pres-ence of many channel errors.[15]. Now thus by implementing this process we getthe required BTC compressed image as the required output. The performancemetrics are measured, tabulated and plotted as graphs respectively.

3.2.4 Enhancement Techniques

Image enhancement is the popular and the most widely known technique of im-age processing. Many images like medical images, satellite images, aerial imagesand even real life photography suffer from noise and poor contrast.Image en-hancement algorithms offer a wide variety of approaches for modifying images toachieve visually acceptable images. The choice of such techniques is a function ofthe specific task, image content, observer characteristics, and viewing conditions.The point processing methods are most primitive, yet essential image processingoperations and are used primarily for contrast enhancement [16]. Enhancementtechniques improve the quality of the image view, blurring, noise and increasingcontrast and improve the borders and sharpness of the image. The enhancementmethods can broadly be divided in to the following two categories,

• Spatial domain

• Frequency domain

Spatial domain and frequency domain include techniques like point processing,image smoothening, edge detection and image sharpening. The techniques usedin this thesis are spatial domain which deal with the image pixels and enhance thecontrast and the compressed medical images are well enhanced by image adjust-ment. The techniques used in thesis to enhance the compressed medical imagesare,

• Adaptive Histogram Equalization (AHE)

• Morphological Operations (MO)

3.2.4.1 Adaptive Histogram Equalization (AHE)

Adaptive Histogram Equalization is the method used for the contrast enhance-ment of images. This is mostly used in gray scale images like medical imageswhere they are in low contrast and they are hence enhanced. The compressed


medical images are enhanced again by using this contrast enhancement methodas it is simple and effective. It generates mapping for each pixels from the sur-rounding windows [10].The method is simple and computationally effective thatmakes it easy to implement and use in real time systems [17].

In this project the medical images from the database that are selected andcompressed are given as an input for enhancement. By using the matlab com-mands and functions the image is enhanced using AHE. Image intensity adjust-ment is also used in combination with AHE so as to enhance the pixels moreclearly. Thus the AHE enhanced output images is obtained for all the lossy andlossless compression techniques. The performance metrics are measured, tabu-lated and plotted as graphs for a clear understanding of the comparisons made.

3.2.5 Morphological Operations (MO)

This technique Morphological Operations (MO) is used in image enhancementof binary images and is also extended to medical images. This is the combina-tion of both erosion and dilation. The images that are compressed are undergonethrough morphological operations where the background of the image is enhancedefficiently using erosion and dilation. Image background approximation to thebackground by means of block analysis in conjunction with transformations thatenhance images with poor lighting. The multibackground notion is introducedby means of the opening by reconstruction shows a comparison among severaltechniques to improve contrast in images Thus the desired enhanced outputs areobtained [18].

3.3 Performance MetricsThe performance metrics that are considered for measuring the compression andenhancement techniques of medical images are as follows,

• Peak Signal to Noise Ratio (PSNR)

• Mean Square Error (MSE)

• Structural Similarity Index Modulation (SSIM)

3.3.1 Peak Signal to Noise Ratio (PSNR)

PSNR is the method which is selected to measure the comparison between thecompression and enhancement techniques. It is the ratio between the maximum


possible power of a signal and the power of a corrupting noise. This performancemetric is the most commonly used as a measure of quality of reconstruction inimage compression and image enhancement.

PSNR = 20 ∗ log10(

255√MSE

)(3.1)

3.3.2 Mean Square Error (MSE)

Mean Square Error (MSE) is another method for comparing the compression andenhancement techniques. This is a criterion for an estimator. MSE minimizes thesum of the squared errors due to bias and variance. The average of the squareof the difference between the desired response and the actual system of the output.

MSE =1

MN

M∑yc=1

N∑xr=1

[I(xr, yc)− I ′(xr, yc)]2, (3.2)

where I(xr, yc) and I ′(xr, yc) are respectively, the original and the recovered pixelvalues at the xr row and yc column for the image of size M×N.

3.3.3 Structural Similarity Index Modulation (SSIM)

Structural similarity Index Modulation is also used to compare the performanceof the image compression and enhancement techniques. This is method which isused to measure the similarity between two images. This method is developed toimprove the techniques like PSNR and MSE. Here, the input compressed imageand the enhanced output image are compared and the structural similarity indexvalue for image I using I’ as the reference image according to,

ssim(I, I ′) =

((2µxµy + C1)(2σxy + C2)

(µ2x + µ2

y+1)(σ2x + σ2

y + C2)

), (3.3)

where C1and C2 are constant and equal to unity and µx, µy, σx, σy and σxyare the local means, standard deviations and cross covariances for the images I,I’.

Chapter 4Results & Analysis

To analyze the results three images are selected and then the specific outputs ofthe respective image are displayed here as results of both compression and en-hancement techniques. The images that are considered are named 111.tif, 222.tifand 333.tif respectively. The values obtained from those medical images are alsotabulated in a tabular form and displayed accordingly.

4.1 Output Image for 111.tifThe Compressed and the Enhanced outputs of this image are displayed one afterthe other below.

• Lossy Techniques

(a) Original Image (b) DCT out compressed image

Figure 4.1.1: Lossy compression of medical image using DCT

14

Chapter 4. Results & Analysis 15

(a) Input Medical Image (b) DWT image after compression

Figure 4.1.2: Lossy compression of medical image using DWT

(a) AHE Enhancement for DCT Com-pressed Image

(b) MO Enhancement for DCT CompressedImage

Figure 4.1.3: Enhancement of DCT compressed image using AHE and MO

(a) AHE Enhancement for DWT Com-pressed Image

(b) MO Enhancement for DWT Com-pressed Image

Figure 4.1.4: Enhancement of DWT compressed image using AHE and MO


• Lossless Techniques

(a) BTC Compressed Image (b) RLE Compressed Image

Figure 4.1.5: Lossless Compression using BTC and RLE

(a) AHE Enhancement for BTC (b) MO Enhancement for BTC

Figure 4.1.6: Enhancement of BTC compressed image using AHE and MO


(a) AHE Enhancement for RLE (b) MO Enhancement for RLE

Figure 4.1.7: Enhancement of RLE compressed image using AHE and MO

4.2 Output Image for 222.tif• Lossy Techniques





















4.3 Output Image for 333.tif• Lossy Techniques






















4.4 Tabular Forms of Performance Parameters

4.4.1 Performance Metrics Tabular Form for 111.tif

Performance metrics PSNR(dB) MSE SSIMoutput image w.r.t inputimageDCT compressed Image 91.94973426 4.15× 10−5 0.986203386AHE enhancement forDCT compressed image

75.37023259 0.00188824 0.963423424

MO enhancement forDCT compressed image

51.64641713 0.445080582 0.050327273

DWT compressed image 91.96046744 4.14× 10−5 0.991260628AHE enhancement forDWT compressed image

75.37304215 0.001887018 0.969275783

MO enhancement forDWT compressed image

51.63273634 0.44648485 0.046785188

Block truncation com-pressed image

79.57898049 0.000716444 0.899149203

AHE enhancement forblock truncation image

74.33496402 0.002396536 0.869922574

MO enhancement forblock truncation image

51.72625772 0.436972986 0.058670346

RLE compressed image 79.50344365 0.000729014 0.90101147AHE enhancement forRLE compressed image

74.44766171 0.002335147 0.87224271

MO enhancement forRLE compressed image

51.72278998 0.437322038 0.058110674

Table 4.1: Image 111.tif tabular form

The performance metrics for image 111.tif are calculated and tabulated for PSNR,MSE and SSIM. For the image 111.tif from the table 4.1 it can be observed MOis the less suitable algorithm to enhance images after compression using lossyand lossless techniques. Comparing PSNR values for MO algorithm that areapproximately 51 dB with greater values in AHE method in the range [74,80] dBshows that MO is less suitable algorithm to enhance images after compression.For this image neither AHE nor MO enhanced the image properly because thePSNR and SSIM values after AHE and MO are lower as compared to PSNR andSSIM values after compression from the table.



Performance metrics PSNR(dB) MSE SSIMOutput images w.r.t in-put imageDCT compressed Image 77.05392243 0.00128141 0.92861073AHE enhancement forDCT compressed image

77.2180706 0.001233881 0.933424602


67.89265671 0.010563667 0.837531458

DWT compressed image 74.30656847 0.002412257 0.467752948AHE enhancement forDWT compressed image

78.91737387 0.00083434 0.926322189


66.26158069 0.015378794 0.837602682


69.39853893 0.007468381 0.834725941


70.4017972 0.005927897 0.796081433


65.03594893 0.020393204 0.746747753


70.22341105 0.006176455 0.767935478


67.07214746 0.012760439 0.74775475


Table 4.2 shows the performance metrics values for the image 222.tif. From thistable we observe that the image with high PSNR value shows good enhancement.AHE enhancement has better PSNR values than compared to MO enhancement.For this image AHE enhanced the image because PSNR values after AHE aregreater (77,78) dB as compared to PSNR values after compression which are (70,72) dB. For RLE, AHE doesn’t enhance much by seeing the values of PSNR forAHE with RLE before (70.20) dB and after enhancement (70.23)dB shows notmuch improvement in PSNR.In this case of DWT compression AHE enhanced the compressed image signif-icantly comparing the PSNR and SSIM values for AHE with DWT before andafter AHE.MO further reduced PSNR’s compared to PSNR’s directly after compression.



Performance metrics PSNR(dB) MSE SSIMOutput images w.r.t in-put imageDCT compressed Image 71.66771337 0.004429034 0.794859736AHE enhancement forDCT compressed image

71.78737069 0.004308671 0.804049513


54.82488614 0.214087715 0.459549559

DWT compressed image 71.40085115 0.004709722 0.736046619AHE enhancement forDWT compressed image

73.57579609 0.002854307 0.879990114


54.37652594 0.237371059 0.497204429


62.27506474 0.038510033 0.524254062


65.37104159 0.018878879 0.513743919


54.20134883 0.247141386 0.328382448


64.99433301 0.02058956 0.522739497


54.95144307 0.207939049 0.359419143


Here, the tabular form is shown for the image 333.tif for PSNR, MSE and SSIMperformance metrics. The SSIM values for DCT are greater than the SSIM valuesfor BTC for the images from the values obtained in the table. For this image AHEenhanced the image because PSNR values after AHE are greater as compared toPSNR values after compression. For RLE, AHE doesn’t enhance much by seeingthe values of PSNR for AHE with RLE before and after.In this case of DWT compression AHE enhanced the compressed image sig-nificantly comparing the PSNR and SSIM values for AHE with DWT before(73,74)dB and after AHE (79,80)dB. MO further reduced PSNR’s to (65,67)dBcompared to PSNR’s directly after compression which are (69,70)dB.


4.5 Plots for PSNR, MSE and SSIM for the imagedatabase

In Figure 4.5.1 high value of PSNR indicates good image quality where these val-ues are high for some images in the 100 images database where the peaks are high.In those places we observe good enhanced images after compression as shown inthe graph plots below.From this plot it is clearly shown that PSNR values shows significant increasewhen using DWT compression and after enhancing it with AHE enhancementtechnique.The PSNR values for MO enhancement are comparatively lower after enhance-ment thus not showing any improvement in the quality.MO enhancement is least suitable for enhancement of medical images because itgives lower PSNR and SSIM values compared to AHE enhancement.


Figure4.5.1:

PSN

Rplot

forallm

edical

images

databa

se.


Figure4.5.2:

SSIM

plot

forallm

edical

images

databa

se.


In Figure 4.5.2 high value of SSIM indicates good image quality where thesevalues are high for some images in the 100 images database where the peaks arehigh. In those places we observe good enhanced images after compression asshown in the graphs plotted .MO enhancement is least suitable for enhancement of medical images because itgives lower PSNR and SSIM values compared to AHE enhancement.The SSIM values for DCT are greater than the SSIM values for BTC for theimages as seen from the plot.

Chapter 5Conclusion & Future Work

The main purpose if this thesis is to analyze the performance parameters and theperformance of the different compression and enhancement techniques. A detailedliterature review has been done to understand the different characteristics andthe working of these techniques. From this literature research, a clear knowledgehas been obtained on the compression and enhancement techniques and how theywork on medical gray scale images.

Firstly, the compression is performed using both lossy and lossless techniquesand then followed by enhancing them. DCT, DWT, RLE and BTC are used forcompression. DWT lossy compression gives better results than DCT when en-hanced based on PSNR, MSE and SSIM without losing more information. RLEand BTC are compress well without loosing much data. RLE shows good com-pression rate than BTC from the analysis.

Each compression technique is further enhanced using AHE and MO tech-niques.Here, we observe the combinations of the compression and enhancementtechniques that worked well together. RLE has good values and better quality ofimages after enhancement rather than BTC by comparing the PSNR and SSIMvalues.The combination of AHE and RLE gives better enhancement results com-pared to any other techniques.

In the case of DWT compression AHE enhanced the compressed image sig-nificantly comparing the PSNR and SSIM values for AHE with DWT before andafter AHE. Morphological operations are used to enhance the background ratherthan the sharpening or increasing the image contrast. this technique in specificis used to enhance the particular region of interest as seen in the results.

There is always a need to explore new methods to find an effective solution.In future, people may also use genetic algorithms and edge detection techniquesand can compare these techniques by using different parameters.

31

References

[1] D. Meenakshi and V. K. Devi, “Literature review of image compression tech-nique,” International Journal of Computer Science & Engineering Technol-ogy, vol. 1, no. 6, pp. 286–288.

[2] S. S. ME, V. Vijayakuymar, and R. Anuja, “A survey on various compressionmethods for medical images,” International Journal of Intelligent Systemsand Applications, vol. 4, no. 3, p. 13, 2012.

[3] S. Bedi and R. Khandelwal, “Various image enhancement techniques-a crit-ical review,” International Journal of Advanced Research in Computer andCommunication Engineering, vol. 2, no. 3, 2013.

[4] K. Cabeen and P. Gent, “Image compression and the discrete cosine trans-form,” College of the Redwoods, 1998.

[5] A. S. Lewis and G. Knowles, “Image compression using the 2-d wavelet trans-form,” IEEE transactions on image processing, vol. 1, no. 2, pp. 244–250,1992.

[6] M. M. H. Chowdhury and A. Khatun, “Image compression using discretewavelet transform,” IJCSI International Journal of Computer Science Issues,vol. 9, no. 4, pp. 327–330, 2012.

[7] S. Benchikh and M. Corinthios, “A hybrid image compression techniquebased on dwt and dct transforms,” 2011.

[8] A. M. A. Ibrahim and M. E. Mustafa, “Comparison between (rle and huff-man) algorithmsfor lossless data compression,” IJITR, vol. 3, no. 1, pp. 1808–1812, 2015.

[9] E. Delp and O. Mitchell, “Image compression using block truncation coding,”IEEE transactions on Communications, vol. 27, no. 9, pp. 1335–1342, 1979.

[10] T. Arici, S. Dikbas, and Y. Altunbasak, “A histogram modification frame-work and its application for image contrast enhancement,” IEEE Transac-tions on image processing, vol. 18, no. 9, pp. 1921–1935, 2009.

32

References 33

[11] P. Maragos, “Morphological filtering for image enhancement and feature de-tection,” analysis, vol. 19, p. 18, 2005.

[12] A. B. Watson, “Image compression using the discrete cosine transform,”Mathematica journal, vol. 4, no. 1, p. 81, 1994.

[13] N. Saroya and P. Kaur, “Analysis of image compression algorithm usingdct and dwt transforms,” International Journal of Advanced Research inComputer Science and Software Engineering, vol. 4, no. 2, 2014.

[14] M. Kamel, C. Sun, and L. Guan, “Image compression by variable blocktruncation coding with optimal threshold,” IEEE Transactions on SignalProcessing, vol. 39, no. 1, pp. 208–212, 1991.

[15] D. Halverson, N. Griswold, and G. Wise, “A generalized block truncationcoding algorithm for image compression,” IEEE transactions on acoustics,speech, and signal processing, vol. 32, no. 3, pp. 664–668, 1984.

[16] R. Maini and H. Aggarwal, “A comprehensive review of image enhancementtechniques,” arXiv preprint arXiv:1003.4053, 2010.

[17] M. Abdullah-Al-Wadud, M. H. Kabir, M. A. A. Dewan, and O. Chae, “A dy-namic histogram equalization for image contrast enhancement,” IEEE Trans-actions on Consumer Electronics, vol. 53, no. 2, 2007.

[18] K. Sreedhar and B. Panlal, “Enhancement of images using morphologicaltransformation,” arXiv preprint arXiv:1203.2514, 2012.

Documents

Comparison of Image Compression and Enhancement …1150997/FULLTEXT02.pdf · data loss is used as the best method for image compression of medical images. Lossless ompression ﬁnds