Contrast Enhancement and Smoothing of CT

Images for Diagnosis

Showkat Hassan Malik Tariq Ahmad Lone S.M.K. Quadri

Department of computer Science Department of computer Science Department of computer Science
Mewar University Mewar University University of Kashmir
Rajasthan India. Rajasthan India Srinagar India
Email Id:shmalik.mca@gmail.com Email Id:tariq380@gmail.com. Email Id:quadrismk@hotmail.com

Abstract— Medical sciences currently rely on the imaging restore images, image smoothing, 2D to 3D conversion and
technology and post processing of these medical images in order to Vice versa using image processing͘
diagnose the diseases. The acquisition of such images may get
affected by inherent noise which can lead to false perception of the
Image Acquisition
diagnostic image. To perceive these images correctly the effective
image enhancement techniques are the foremost requirement to
improve the quality and level of perception. In this paper we
apply various image enhancement techniques on CT images. Imaging Hardware
Resultant images are inspected visually by radiologists for
performance evaluation. Median filter removes noise and
smoothen CT images effectively and outperforms other Filtering Image Processing
techniques. While as adaptive histogram equalization enhances
the contrast and outperforms other contrast enhancement
techniques.
Image Display
Keywords: - Digital image processing, CT images, Noise, Image
enhancement, Medical imaging processing.
Image Reading Environment
I. INTRODUCTION

Diagnosis of medical images depends on image acquisition, Diagnosis

image processing and image display [1]. Medical images can
get affected by various kinds of noise during image acquisition
processes. These images are not adequate for the diagnostic
Fig. 1 Diagnostic Process steps
process. Image enhancement techniques can effectively be used
to remove noise and enhance the quality of input images so that Now it depends also on the imaging technology that we are
they are used for diagnosis of various diseases. Removal of using for image acquisition what can be the type of inherent
inherent noise from a medical image is a pre-processing step in noise. We can choose filtering techniques accordingly in order
medical image processing. There are various filtering methods to remove the noise and smooth the input image. Image
used to remove noise from these images like mean filter etc. acquisition technology also derives the type of image
Low contrast images are also not adequate for the diagnosis of enhancement technique that we should use. We discuss
diseases. Contrast enhancement is extensively used in medical filtering and contrast enhancement techniques and then we
images processing. Contrast enhancement is also used as a pre- apply these techniques on CT images. The resultant images and
processing step in medical image processing [2]. There are image Histogram to show the distribution of pixel intensity
many techniques for contrast enhancement like classical values graphically are shown in figures below.
histogram equalization etc. The paper is organized as follows: Section II presents the
The steps followed in diagnosis of medical images are shown filtering techniques. Section III presents contrast enhancement
in the diagram above. As shown in the figure 1 image techniques. Section IV presents Image quality measures.
processing role comes after image acquisition. We can enhance Section V presents experimental results and discussion. Section
correct colour, detect contours, calculate area of cells, and VI concludes the paper.

978-9-3805-4416-8/15/$31.00 2015
c IEEE 2214
 II. FILTERING TECHNIQUES.
Filtering techniques can be effectively used to remove inherent
noise from medical images [3]. Filtering techniques transform
pixel intensity values to reveal certain image characteristics.
Enhancement
Smoothing and Template Matching
Among the above we deal with enhancement and Smoothing of
medical images. The paper will discuss some of the popular
filtering techniques for noise removal and image smoothing.
A. Mean filter.
Mean filter or average filter comes in the classification of
windowed filter of linear class which are used to remove noise
and smoothing images. The concept works on averaging the
neighbours of any element of the image. The average filter
computes the sum of all pixels in the filter window and divides
it by total number of pixels [3]. Then it replaces centre pixel
value with the computed average. Result value for any pixel
Fig. 3 Median Filter
with index (i, j) will be computed as follows
C. Weighted median filter
ሺ‹ǡ Œሻ ൌ ሾሺ‹ െ ͳǡ Œ െ ͳሻ ൅ ሺ‹ െ ͳǡ Œሻ ൅ ሺ‹ െ ͳǡ Œ ൅ ͳሻ This filter has edge preserving capacity of the classical median
൅ ሺ‹ǡ Œ െ ͳሻ ൅ ሺ‹ǡ Œሻ ൅ ሺ‹ǡ Œ ൅ ͳሻ filter. Filter is yielded by adjusting weights. Smallest and
൅ ሺ‹ ൅ ͳǡ Œ െ ͳሻ ൅ ሺ‹ ൅ ͳǡ Œሻ largest values are ignored. Result image values are computed
൅ ሺ‹ ൅ ͳǡ Œ ൅ ͳሻሿȀͻሺͳሻ by sorting all the pixels in the window. Let x1” x2”.... ” xn be
sorted values. Let a1, a2 ... an be corresponding weights. Then
we compute median by considering weights of corresponding
pixels. Resultant computed value will replace the original pixel
value [5].

Fig..2 Mean Filter

B. Median filter
Median filter is a non linear filtering technique also used to Fig. 4 Weighted Median Filter
remove noise and smoothing images unlike mean filter it
preserves useful details in the image. The concept of median D. Wiener Filter
filter works on by replacing the pixel value with the median of The Wiener filter is the MSE optimal linear filter for blurred
its neighbouring pixels [4]. Result image value for any pixel images and images contaminated with additive noise. This
with index (i, j) will be computed by sorting all the pixels in the filter requires the information about the spectra of the noise and
window that are pixels with index (i-1, j-1), (i-1, j), (i-1, j+1), (i, the original signal and works well only if the underlying signal
j-1), (i, j), (i, j+1), (i+1, j-1), (i+1, j) and (i+1, j+1). Then is smooth. Suppose G (u, v) is the filter chosen then the
calculate the median of these values. The resultant value will resultant image is given by
replace the original pixel with index (i, j). ܺሺ‫ݑ‬ǡ ‫ݒ‬ሻ ൌ ‫ܩ‬ሺ‫ݑ‬ǡ ‫ݒ‬ሻǤ ܻሺ‫ݑ‬ǡ ‫ݒ‬ሻሺʹሻ

2015 2nd International Conference on Computing for Sustainable Global Development (INDIACom) 2215
 We choose
ሺ—ǡ ˜ሻ so that mean square is minimized [6].

Fig. 6 Histogram Equalization

Fig. 5 Wiener Filter
B. Adaptive histogram equalization
III. CONTRAST ENHANCEMENT TECHNIQUES Adaptive histogram equalization is a technique used for
Contrast enhancement techniques are used to enhance the contrast improvement. It computes several histograms
contrast of low contrast images effectively. The aim of contrast corresponding to distinct image sections and then uses these
enhancement is to differentiate between various objects in an sections of images to redistribute brightness value of the image.
image and increase the perception of information from an It is suitable for low contrast images for improving the
image. There are various contrast enhancement techniques. The perception of images [8] [12].
paper will discuss some of the popular contrast enhancement
techniques.
A. Histogram equalization
Histogram equalization (HE) is a technique for adjusting image
intensities to enhance image contrast. Histogram of an image
represents the frequency of occurrence of all the levels in the
image [7][11].
Let f be a given image represented as a matrix of integer pixel
intensities ranging from 0 to L-1. L is the total number of
possible intensity values, often 256 [10]. Let p denote the
normalized histogram of f with a bin for each possible intensity.

—„‡”‘ˆ’‹š‡Ž•™‹–Š‹–‡•‹–›
’୬ ൌ ሺ͵ሻ
–‘–ƒŽ—„‡”‘ˆ’‹š‡Ž•
Fig. 7 Adaptive Histogram Equalization
 ൌ Ͳǡͳǡʹǡ ǥ ǡ  െ ͳ
C. Contrast stretching
The histogram equalized image will be defined by The technique Contrast stretching is mainly used for
௙೔ǡೕ
improving images by stretching the range of intensity values it
݃௜ǡ௝ ൌ ݂݈‫ ݎ݋݋‬൮ሺ‫ ܮ‬െ ͳሻ ෍ ܲ௡ ൲ሺͶሻ contains to utilize the possible values. It performs a linear
௡ୀ଴
mapping of input values to output values.
Firstly the limits over which the image intensity values will be
Where floor(x) rounds to the greatest integer less or equal to stretched are determined. Let these lower and upper limits be a
x [14] [10]. and b, respectively. Next, the histogram of the original image is
analysed to determine the limit values (lower limit=c, upper
limit=d) in the output image [9] [10] [11].The value (r) in the
original image is mapped to the value(s) in the output image
using the following function:

2216 2015 2nd International Conference on Computing for Sustainable Global Development (INDIACom)
 ሺ„ െ ƒሻ ‫ܺܣܯ‬௜ଶ
• ൌ ሺ” െ …ሻ ቆ ቇ ൅ ƒሺͷሻ ܴܲܵܰ ൌ ͳͲǤ ݈‫݃݋‬ଵ଴ ቆ ቇሺ͹ሻ
ሺ† െ …ሻ ‫ܧܵܯ‬

Where ‹ is the maximum possible pixel intensity

value, when pixel is represented by 8 bits its value is 255.

C. Root Mean Square Error

Root mean square error is calculated by taking the square root
of the MSE discussed above is given by
୑ିଵ
ͳ ୒ିଵ
 ൌ ඩ ෍ ෍ ሾሺ‹ǡ Œሻ െ ሺ‹ǡ Œሻሿଶ ሺͺሻ
 ୨ୀ଴
୧ୀ଴

Where matrix I represents input image and matrix K represents

output image. Low is the value of RMSE more is the quality of
enhanced image. Greater is the value low is the quality [6].

D. Universal image quality Index

Fig. 8 Contrast Stretching Universal image quality index uses combination of three
factors to evaluate image quality that is Loss of correlation,
IV IMAGE QUALITY MEASURES luminance distortion and contrast distortion. The value of
We use quantitative methods and visual inspection of Universal image quality index denoted by Q is given by
radiologists for image quality evaluation and comparison. It is
difficult to quantify an enhancement of image perception. ɐ୶୷ ʹšӯ›ӯ ʹɐ୶ ɐ୷
Desirably we have used peak signal to noise ratio (PSNR), ൌ Ǥ ଶ Ǥ ଶ ሺͻሻ
ଶ
ɐ୶ ɐ୷ šӯ ൅ ›ӯ ɐ୶ ൅ ɐଶ୷
Mean Square Error (MSE), Root Mean Square Error (RMSE),
Universal Image Quality Index (UIQI), Enhancement
Where ɐ୶ is variance of x is, ɐ୷ is variance of y and ɐ୶୷ is
Measurement Error (EME) and Pearson correlation coefficient
as quantitative measures. Results of these techniques are shown covariance of x, y. The first component is the correlation x and
in Table1 and Table2. y, second component is the luminance between x and y and the
third contrast similarity between x and y. the value of Q is in
A. Mean Square error (MSE) the range [0, 1] [15].

MSE measures the difference between values of original image E. Pearson correlation coefficient
and the resultant image. MSE is a function corresponding to the
expected values of the squared error. MSE measures the average It is a quantity denoted by r used to compute the quality of least
of the squares of the errors. The error is the difference between square fitting to the original image/data and is given by
ଶ
Expected value and the estimated value. Lower MSE generally ••୶୷
”ൌ ሺͳͲሻ
indicates that the output image is of higher quality [6]. ••୶୶ ••୷୷
ெିଵ
ͳ ேିଵ Where
‫ ܧܵܯ‬ൌ ෍ ෍ ሾ‫ܫ‬ሺ݅ǡ ݆ሻ െ ‫ܭ‬ሺ݅ǡ ݆ሻሿଶ ሺ͸ሻ ୬
••୶୶ ൌ ෌୧ୀଵሺš୧ െ šӯሻଶ is variance of x
‫ܰܯ‬ ௝ୀ଴
௜ୀ଴ ୬
••୷୷ ൌ ෌୧ୀଵሺ›୧ െ ›ӯሻଶ is variance of y.
Where matrix I represents input image and matrix K ••୶୷ ൌ σ୬୧ୀଵሺš୧ െ šӯሻ ሺ›‹ െ ›ӯሻ is covariance of x and y
represents output image. [16].

B. Peak Signal to Noise Ratio (PSNR) F. Enhancement measurement error

PSNR is used to measure the quality of processed image of the Enhancement measurement error (EME) divides the input
original image. PSNR is expressed in terms of logarithmic image into n1n2 non-overlapping sub blocks. The EME value
decibel scale [6] [10]. Higher PSNR generally indicates that the is computed as follows
output image is of higher quality. PSNR is defined in terms of ୬ଵ
MSE as: ͳ ୬ଵ ƒšሺš୧ ǡ୨ ሻ
 ൌ ෍ ෍ ʹͲŽ ሺͳͳሻ
ଵ ଶ ୨ୀଵ ‹ሺš୧ ǡ୨ ሻ
୧ୀଵ

2015 2nd International Conference on Computing for Sustainable Global Development (INDIACom) 2217
 After applying the mentioned filtering and contrast
Where max (xi,j) and min (xi,j) are the maximum and minimum enhancement techniques the results shown in figures 2 to 8
grey levels in block xi,j respectively. Greater the value of EME above reveals image smoothing, noise removal and contrast
more is the quality of input image [17]. enhancement. According to visual inspection by radiologists
Mean filter makes CT scan images smoother but does not
TABLE I. VALUES OF IMAGE QUALITY MEASUREMENT TECHNIQUES AFTER
preserve image details well for further processing. Median
APPLYING FILTERING TECHNIQUES filter effectively removes Gaussian noise, makes images
smoother and preserves images details. Wiener filter removes
Evaluation Mean Median Weighted Wiener Gaussian noise from CT scan images and also preserves image
Technique Filter Filter Median Filter
details for further processing. While as according to image
Filter
quality measuring techniques median filter outperforms other
PSNR(db) 23.7819 26.85377 23.74056 25.84279 filtering techniques as far as PSNR, MSE, RMSE and UIQI are
8 concerned shown in Table1 above. Wiener filter also performs
MSE 274.335 135.23940 276.96460 170.68723 well as per Pearson correlation coefficient and EME.
28 Contrast enhancement techniques are used to enhance contrast
16.5651 16.64225 13.66473
of CT scan images in order to differentiate between various
RMSE 11.62925
6 organ parts. The simulation results obtained after applying
UIQI 0.29318 0.36900 0.29202 0.35761 contrast enhancement techniques are shown in figures 6 to 8
above. According to visual inspection Adaptive histogram
Pearson 63872.8 64508.236 63871.623 64649.84931
correlation 5936 73 06
equalization enhances contrast of CT scan images effectively
coefficient than that of histogram equalization and contrast stretching. It
also preserves image details for further processing. While as
EME 5.02067 5.02067 5.02067 5.02067 according to image quality measuring techniques contrast
(Original stretching outperforms other techniques as far as PSNR, MSE,
Image) RMSE, UIQI and Pearson correlation coefficient are
concerned. While as ADE performs well as per EME shown in
EME 24.3708 5.61441 23.88819 27.86596
8
Table2 above.
(Enhanced
Image)
VI. CONCLUSION.
The paper discusses popular filtering and contrast enhancement
techniques for noise removal, smoothing, and contrast
TABLE II. VALUES OF IMAGE QUALITY MEASUREMENT TECHNIQUES AFTER enhancement of CT scan images. After acquiring results from
APPLYING CONTRAST ENHANCEMENT TECHNIQUES experiment as per visual inspection we find that median filter
Evaluation Histogram Adaptive Contrast performs better in order to remove Gaussian noise and make
Technique Equalization Histogram Stretching them smooth. As per image quality measuring techniques
Equalization median filter outperforms other techniques. While as as per
PSNR(db) 12.99710 16.46668 33.12203 visual inspection in contrast enhancement techniques AHE
MSE 3286.77538 1478.46776 31.93576 technique improves the low contrast CT scan image and also
RMSE 57.33041 38.45085 5.65117 preserves images details for Diagnosis and as per image quality
UIQI 0.65288 0.68414 0.95104 measuring techniques contrast stretching outperforms other
Pearson 44398.21437 44398.21437 48141.32248 techniques. Hence in filtering techniques median filter
correlation outperforms other filtering techniques. Adaptive histogram
coefficient
equalization and contrast stretching performs well in contrast
EME 24.99584 24.99584 24.99584
(Original
enhancement techniques as far as CT images of brain are
Image) concerned.
EME 22.75274 37.95058 20.40270
(Enhanced VII. REFERENCES
Image)
[1] E. A. Krupinski, Current perspectives in medical image
perception, Attention, Perception & Psychophysics. 72(5).
V. EXPERIMENTAL RESULTS AND DISCUSSION
2010.
The experiment was carried out on matlab R2009 and the source [2] Castleman. Kenneth. R, Digital Image Processing, Prentice
images were taken from the database maintained by computer Hall, New Jersey, 1979.
vision group. The images provided by this group are copy right [3] R. C. Gonzales and R. E. Woods, Digital Image
free [*].We have chosen CT scan images of brain for our Processing, Prentice Hall, New Jersey 2008.
experiment.

2218 2015 2nd International Conference on Computing for Sustainable Global Development (INDIACom)
[4] S. P. Windyga, Fast Impulsive Noise Removal, IEEE
transactions on image processing, Vol. 10, No. 1, pp. 173-178,
2001.
[5] Lin Yin ; Ruikang Yang ; Gabbouj, M. ; Neuvo, Y.Circuits and
Systems II: Analog and Digital Signal Processing, IEEE
Transactions on (Volume: 43, Issue: 3) Mar 1996.
[6] A. K. Jain, Fundamentals of Digital Image Processing, Prentice
Hall, 1989.
[7] R. Hummel, “Histogram modification techniques“, Computer
Graphics and Image Processing, Vol. 4, pp. 209-224, 1975.
[8] Pizer, S.M.; Johnston, R.E.; Ericksen, J.P.; Yankaskas, B.C.;
Muller, K.E., "Contrast-limited adaptive histogram
equalization: speed and effectiveness," Visualization in
Biomedical Computing, 1990., Proceedings of the First
Conference on , vol., no., pp.337,345, 22-25 May 1990.
[9] Ms.Seema Rajput, Prof.S.R.Suralkar, “Comparative Study of
image Enhancement Techniques”, IJCMC, vol. 2, Issue 1,
Pg.11-21, Jan 2013.
[10] Malik, S.H.; Lone, T.A., "Comparative study of digital image
enhancement approaches," Computer Communication and
Informatics (ICCCI), 2014 International Conference on , vol.,
no., pp.1,5, 3-5 Jan. 2014.
[11] S.E. Umbagh, Computer Vision and Image Processing,
Prentice Hall, New Jersey, 1998.
[12] Stark, J.A., "Adaptive image contrast enhancement using
generalizations of histogram equalization," Image Processing,
IEEE Transactions on , vol.9, no.5, pp.889,896, May 2000.
[13] http://decsai.ugr.cs/index2 retrieved on Aug 2014.
[14] sukhjinder singh, et al., “comparative study and
implementation of image processing techniques using matlab”,
international journal of advanced research in computer science
and software engineering, vol. 2, issue 3, March 2012.
[15] Zhou Wang; Bovik, A.C., "A universal image quality
index," Signal Processing Letters, IEEE , vol.9, no.3, pp.81,84,
March 2002.
[16] Panetta K., Zhou Y., Agaian S., and Jia H, “Nonlinear Unsharp
Masking for Mammogram Enhancement”, IEEE Transactions
on Information Technology in Biomedicine, vol. 15, no. 6, pp.
918-928, 2011.
[17] Huang, S. C., Cheng, F. C., & Chiu, Y. S. (2013). Efficient
contrast enhancement using adaptive gamma correction with
weighting distribution.Image Processing, IEEE Transactions
on, 22(3), 1032-1041.

2015 2nd International Conference on Computing for Sustainable Global Development (INDIACom) 2219