International Journal of Engineering Trends and Technology (IJETT) – Volume 9 Number 9 - Mar 2014
A Research on Various Filtering Techniques in
Enhancing Mammogram Image Segmentation
Cholavendhan Selvaraj#1, Siva Kumar R*2, Karnan M#3
#
Dept. of Computer Science & Engineering, Tamilnadu College of Engineering, Coimbatore, India
*Dept. of Information Technology, Tamilnadu College of Engineering, Coimbatore, India
Abstract- Breast cancer is the most common death causing
cancer among women. Finding an efficient and accurate
segmentation techniques still remains a challenging problem in
digital mammography. Pre-processing of images play a vital role
in efficient segmentation because of several factors which affects
the efficiency and accuracy of further processing. This paper
aims to study various available pre-processing approaches and
finds the best suitable existing approach for enhancing medical
mammogram images for better representation. Performance of
the filters are evaluated using Peak Signal to Noise ratio and
Mean Square Error approaches and compared against each
other for different noises.
Keywords- Filter, Enhancement, Pre-processing, Mammography
and Breast Cancer.
A. Noise Removal
Images acquired from the real world are subjected to various
types of noises. Noises such as marks and lines are present in the
majority of acquired mammogram images. For effective processing
these noises should be removed before processing the image. This
can be accomplished by using filters. Filters like mean, median,
average filter, wiener filter, spatial low pass filter, Gaussian filter,
bilateral filter, Butterworth filter and wavelet filters are applied to
noise image to compare and find which filter yields the best quality
to the original image.
For example Salt and pepper noised image as shown in
Figure 1.b) is filtered using a 2-D Median filtering approach using a
3-by-3 neighbourhood connection and Spatial Filtering. Each output
pixel in filtered image contains median value of neighbourhood to
the corresponding pixel in the input images as shown in figure 1c.
I. INTRODUCTION
The early detection of breast cancer can be key to survival.
Breast self-examinations, clinical breast exams by the experts, and
screening mammography are essential methods which are used to
detect the cancer in breasts at an early stage. There are a several
factors associated with effective mammogram segmentation. Super
imposition of several types of tissues in the breast region makes it
very difficult to differentiate and identify the regions. Several steps
are involved in efficient segmentation of mammogram image. It
involves mammogram image pre-processing, enhancement and
segmentation.
Digital mammogram images are acquired from mini MIAS
database. The images are digitized at 200 micron pixels edge and
padded in order to obtain all images with a size of 1024 × 1024
pixels as shown in figure 1.a).
Figure 1 a) Original Image b) Salt & Pepper Noised image c) Median
Filtered image
II.IMAGE ENHANCEMENT
Mammogram image enhancement is the process of
manipulation of images by reducing noises and increase the image
contrast in order to detect the abnormalities. The methods used to
manipulate mammogram images can be categorized into four main
categories namely the conventional enhancement techniques, the
region-based, feature-based and fuzzy enhancement techniques.
Conventional enhancement techniques used to modify the
mammogram images based on the global properties as it is a fixed
neighbourhood technique. Region-based techniques are used for
contrast enhancement of mammogram images in accordance to its
surroundings. Feature based methods are based on wavelet domain
enhancement techniques. The fuzzy enhancement techniques apply
fuzzy operators and properties to enhance the mammogram image.
ISSN: 2231-5381
d) Spatial low Pass Filtered Image e) Wiener Filter f) Bilateral Filter
http://www.ijettjournal.org
Page 451
International Journal of Engineering Trends and Technology (IJETT) – Volume 9 Number 9 - Mar 2014
The noises are added to the original image manually of
various intensities. Then filters are applied to those noises and results
are compared against each other to find the better one.
A. Mean Square Error
The mean-square error is an average or expected value of
squared error or loss. It can be calculated by using the following
equation:
∑
Figure 2 Resultant image filtered from Gaussian noisy image a) Gaussian
Noise b) Spatial Low pass Filter c) Average Filtered Image
(1)
M and N represents the number of rows and columns in the
input images I1 and I2.
B. Peak Signal Noise Ratio
One main disadvantage of MSE is that it depends strongly
on the image intensity scaling. Peak Signal-to-Noise Ratio (PSNR)
avoids this problem by scaling the MSE according to the image
range:
(2)
where R is the maximum pixel value. PSNR is measured in decibels
(dB). The measure of the signal strength by means of square is the
main disadvantage of PSNR.
Figure 4 a) Image with Speckle Noise b) Spatial Low pass Filtered image
c) Bilateral Filtered image
IV.EXPERIMENTAL RESULTS
The performance of various filters applied to noises are
evaluated and compared in terms of Peak Signal Noise ratio and
Mean Square Error. The values are tabulated as shown in Table I &
II and Figure 5.
TABLE I
MEAN SQUARE ERROR
Filter
Figure 4 a) Speckle Noised image b) Spatial Low pass Filtered image
c) Average Filtered image
Spatial
Average Gaussian Bilateral
Median Wiener
Low
Filter
Filter
Filter
Filter Filter
pass
Poisson
Noise
Gaussian
Noise
Salt
&
Pepper
Speckle
Noise
0.0002
0.0003
0.0002 0.0001 0.0002 0.0003
0.0015
0.0035
0.0026 0.0015 0.0013 0.0023
0.0026
0.0083
0.0189 0.0026
0.0006
0.0021
0.0026 0.0006 0.0014 0.0024
0
0.0127
III.PERFORMANCE ANALYSIS
Aim of this paper is to find the filter which performs best
in extracting the image from noisy mammogram image. In order to
find the effective one, performance of the filters are analysed and
compared against each other. Resulted extracted images from noisy
images are compared with original images. This involves number of
steps like
1. Load the Original input mammogram image.
2. Add noise manually to evaluate the performance of filers to
various noises
3. Apply various filters to different noises.
4. Calculate Mean Square Error and Peek Signal Noise Ratio
5. Compare the performance of various filter against each other
for various noises added to original images.
ISSN: 2231-5381
TABLE II
PEAK SIGNAL NOISE RATIO
Noise/Filter
Average
Filter
Gaussi
Spatial
Bilateral
Median Wiener
an
Low Pass
Filter
Filter
Filter
Filter
Filter
Poisson
85.232 78.761
Noise
Gaussian
64.088 56.062
Noise
Salt
&
57.337 47.047
Pepper
Speckle
73.340 61.711
Noise
http://www.ijettjournal.org
84.673
86.6718 86.216 80.336
58.362
64.2304 66.111 60.624
39.667
57.3206 109.242 43.104
58.959
73.715
65.507 60.125
Page 452
International Journal of Engineering Trends and Technology (IJETT) – Volume 9 Number 9 - Mar 2014
Poisson Noise
Gaussian Noise
Salt & Pepper Noise
Speckle Noise
REFERENCES
120
100
80
60
40
20
0
Average Gaussian Bilateral Spatial Median Wiener
Filter
Filter
Filter Low pass Filter
Filter
Figure 5 PSNR of filtered image
V.CONCLUSION
Several filters have been applied to different type of noises
with varying levels. The performance of those filters are calculated
and plotted in terms of peak signal noise ratio and mean square error.
Among all the filters have been experimented, median and spatial
low pass filter performs well against noises. Spatial filters yields
better outcome for images with Poisson and Speckle noise. Median
filter performs well against Gaussian noise and its best against Salt
and Pepper noise images. Images can be further enhanced using
contrast adjustment and histogram equalization methods to make it
easier and efficient for processing.
ISSN: 2231-5381
[1]. Armen Sahakyan and Hakop Sarukhanyan, “Segmentation of the
Breast Region in Digital Mammograms and Detection of Masses”,
International Journal of Advanced Computer Science and
Applications, Page no. 102, Vol. 3, No.2, 2012.
[2]. Thangavel K, Karnan M, Sivakumar R, KajaMohideen A, “Ant
Colony System for Segmentation and Classification of
Microcalcification in Mammograms”, International Journal on
Artificial Intelligence and Machine Learning, Vol 5, Issue 3, Pages
29-40, 2005.
[3]. Sahakyan, “Segmentation of Mammography Images Enhanced by
Histogram Equalization”, Mathematical Problems of Computer
Science 35, Armenia, pp. 109 - 115, 2011.
[4]. Jonas Krause, Jelson Cordeiro, Rafael Stubs and Heitor Silverio
Lopes, “A Survey of Swarm Algorithms Applied to Discrete
Optimization Problems”, SIBIC, 2013.
[5]. Akanksha Sharma and Parminder Kaur, “Review of CAD
Techniques for Liver Tumor Detection”, International Journal of
Advanced Research in Computer Science and Software Engineering,
Volume 3, Issue 10, October 2013.
[6]. Siddhartha Bhattacharyya, “A Brief Survey of Color Image
Preprocessing and Segmentation Techniques”, Journal of Pattern
Recognition Research, pp.120-129, 2011.
[7]. R. Chandrasekhar, and Y. Attikiouzel, Y. Automatic, “Breast Border
Segmentation by Background Modeling and Subtraction”,
Proceedings of the 5th International Workshop on Digital
Mammography (IWDM), Medical Physics Publishing, pp. 560–565,
2000.
[8]. T. Huang, G. Yang, G. Tang, “A fast two-dimensional median
filtering algorithm”, Acoustics, Speech and Signal Processing, IEEE
Transactions on, vol. 27, 1979.
[9]. www.mathworks.in/help/vision/ref/psnr.html
http://www.ijettjournal.org
Page 453