International Journal of Scientific Research in Computer Science, Engineering and Information Technology

© 2018 IJSRCSEIT | Volume 3 | Issue 3 | ISSN : 2456-3307

Analysis of Image Quality using LANDSAT 7 ETM+ and Gaussian

Filter
A. L. Choodarathnakara*1 and Sinchana G S2
*1Assistant Professor, Dept. of Electronics & Communication Engineering, Government Engineering College,
Kushalnagar, Karnataka, India
2 UG Scholar, USN: 4GL15EC045, Department of E&C Engineering, GEC, Kushalnagar, Karnataka, India

ABSTRACT

Interpretation of image contents is a significant objective in computer vision and image processing. An image
contains different information of scene, such as objects shape, size, color and orientation, but discrimination of
the objects from their background is the first essential task that should be performed before any interpretation.
Filters re-evaluate the value of every pixel in an image. For a particular pixel, the new value is based on pixel
values in a local neighborhood, a window centered on that pixel, in order to reduce noise by smoothing and
enhance edges. At the same time as reducing the noise in a signal, it is important to preserve the edges. Edges
are of critical importance to the visual appearance of images. So, it is desirable to preserve important features,
such as edges, corners and other sharp structures, during the denoising process. In this paper an attempt is
made to assess the impact of bandwidth on image quality using Gaussian filter and LANDSAT 7 ETM+ satellite
imagery. The study area considered for the experimentation is the Mysore city in the state of Katnataka. From
the experimental observations, for a satellite image with high resolution around 30m, the window 5x5 is
recommended which improves image while preserving the edges.
Keywords : Remote Sensing, Image Processing, Gaussian Filter, LANDSAT 7 ETM+, Matlab.

I. INTRODUCTION are developed based on some physiological

observations and important properties of the
The filtering methods are used for image Gaussian function that enable to perform edge
enhancement and denoising. There many types of analysis in the scale space. Marr and Hildreth were
filters are used to enhance an Image. Median filter it the pioneers that proposed an edge detector based on
would be removing noise in image, but the Gaussian filter. Their method had been a very
median filter it depends on mask size chosen. popular one, before Sobel released his detector. They
The mask size is larger, then the image become originally pointed out the fact that the variation of
blurring and over smoothing. Image denoising is also image intensity (i.e. edge) occurs at different levels.
called image restoration techniques. This implied the demand to smoothing filters with
different scales, since a single filter cannot be optimal
Gaussian filters are the most widely used filters in for all possible levels. They suggested the 2D
image processing and extremely useful as detectors Gaussian function, defined as following, as the
for edge detection. It is proven that they play a smoothing operator.
significant role in biological vision particularly in
human vision system. Gaussian based edge detectors
…… (1)

CSEIT1833473 | Received : 20 March 2018 | Accepted : 10 April 2018 | March-April-2018 [ (3 ) 3 : 1296-1304 ] 1296
 A. L. Choodarathnakara et al. Int J S Res CSE & IT. 2018 Mar-Apr;3(3) : 1296-1304

Where, is the standard deviation, and (x, y) are the optical scanners of the satellite like Landsat MSS, IRS
Cartesian coordinates of the image pixels. They etc. In digital image, the data is arranged in a matrix
showed that by applying Gaussian filters of different of a number or brightness value and is located in X,
scales i.e., to an image; a set of images with Y-coordinate system. A pixel is “a two dimensional
different levels of smoothing will be obtained. picture element that is the smallest non-divisible
element of a digital image”. The dataset may be of
There are advantages for the Gaussian filters that multispectral bands and each pixel has brightness
make it unique and so important in edge detection. value or digital number from each of these
The first concerns to its output. It is proven that multispectral bands. The digital numbers
when an image is smoothed by Gaussian filter, the representing the variation in the intensity of radiant
existing zero-crossings (i.e. detected edges) disappear energy emitted from different earth surface features.
as moving from fine-to-coarse scale, but new ones The digital number starts from 0 and goes up to a
are never created. higher value on the gray scale, for example, 0 to 127
in IRS series. Lower the brightness value in the gray
This unique property makes it possible to track zero- scale, lower the intensity of radiant energy from that
crossings (i.e. edges) over a range of scales, and also area of the ground, higher the value, higher the
gives the ability to recover them at sufficiently small intensity of the radiant energy of the area of the
scales. Yuille and Poggio proved that with the ground. In 7-bit data, 0 represents black and 127
Laplacian, the Gaussian function is the only filter in a represents white.
wide category that does not create zero-crossings as
the scale increases. The dual tree complex wavelet The first step towards designing an image analysis
transform having the high computational system is digital image acquisition using sensors in
complexity is high compared with wavelet. There optical or thermal wavelengths. A two dimensional
are iterative algorithms also used for image denoising. image that is recorded by these sensors is the
There are landweber, total variation and blinded- mapping of the three-dimensional visual world. The
convolution methods are resulting good output, but captured two dimensional signals are sampled and
produces moderated quality measure. quantized to yield digital images. Sometimes we
receive noisy images that are degraded by some
In this paper an attempt is made to assess the impact degrading mechanism. One common source of image
of bandwidth on image quality using Gaussian filter degradation is the optical lens system in a digital
and LANDSAT 7 ETM+ satellite imagery. The study camera that acquires the visual information. If the
area considered for the experimentation is the camera is not appropriately focused then we get
Mysore city in the state of Katnataka. From the blurred images. Here the blurring mechanism is the
experimental observations, for a satellite image with defocused camera. In some cases there may be a
high resolution around 30m, the window 5x5 is relative motion between the object and the camera.
recommended which improves image while Thus if the camera is given an impulsive
preserving the edges. displacement during the image capturing interval
while the object is static, the resulting image will
II. DIGITAL IMAGE PROCESSING invariably be blurred and noisy. In some of the above
cases, we need appropriate techniques of refining the
Digital Image is a representation of an object, which images so that the resultant images are better visual
comprised of discrete picture elements (Pixel), each quality, free from aberrations and noises. Hence,
having a digital number generated by the electro- image enhancement, filtering, and restoration have

Volume 3, Issue 3 | March-April-2018 | http:// ijsrcseit.com 1297

 A. L. Choodarathnakara et al. Int J S Res CSE & IT. 2018 Mar-Apr;3(3) : 1296-1304

been some of the important applications of image Kaveri and Kabini rivers. The city got its first piped
processing since the early days of the field. water supply when the Belagola project was
commissioned in 1896. As of 2011, Mysore gets
III. STUDY AREA AND METHODOLOGY 42.5 million gallons water per day. Mysore
sometimes faces water crises, mainly during the
A. Study Area summer months (March–June) and in years of low
The Mysore, officially renamed as Mysuru, is the rainfall. The city has had an underground drainage
third most populous city in the state of Karnataka, system since 1904. The entire sewage from the city
India located at the base of the Chamundi Hills about drains into four valleys: Kesare, Malalavadi, Dalavai
146 km (91 mi) southwest of the state capital and Belavatha. In an exercise carried out by the
Bangalore. It is spread across an area of 128.42 km2 Urban Development Ministry under the national
(50 sq mi) & Mysore is located at 12.30°N 74.65°E and urban sanitation policy, Mysore was rated the second
has an average altitude of 770 meters (2,526 ft). cleanest city in India in 2010 and the cleanest in
People in and around Mysore extensively use Karnataka.
Kannada as medium of language. Mysore has several
lakes, such as the Kukkarahalli, the Karanji, and the
Lingambudhi lakes. In 2001, total land area usage in
Mysore city was 39.9% residential, 16.1% roads,
13.74% parks and open spaces, 13.48% industrial,
8.96% public property, 3.02% commercial, 2.27%
agriculture and 2.02 water. The city is located
between two rivers: the Kaveri River that flows
through the north of the city and the Kabini River, a
tributary of the Kaveri that lies to the south.
According to the provisional results of the 2011
national census of India, the population is 887,446.

Urban growth and expansion is managed by the Figure 1. Google Earth Snapshot of Mysore Study
Mysore Urban Development Authority (MUDA), Area
which is headed by a commissioner. Its activities B. Data Products
include developing new layouts and roads, town
The Table I provide the specification of satellite
planning and land acquisition. One of the major
image data products used in this work. Another data
projects undertaken by MUDA is the creation of an
used is the visual band multi spectral data
Outer Ring Road to ease traffic congestion. Mysore
downloaded from www.wikimapia.com. The Table
has a tropical savanna climate designated under the
II specifies the characteristics of LANDSAT 7 ETM+
Koppen climate classification. The main seasons are
band with ground features.
summer from March to June, the monsoon season
Table I Details of Satellite Data Products Used
from July to November and winter from December
to February. The highest temperature recorded in
Mysore was 39.4 °C (103 °F) on 4 April 1914, and the
lowest was 7.7 °C (46 °F) on 16 January 2012. The
city's average annual rainfall is 804.2 mm (31.7 in).
Drinking water for Mysore is sourced from the

Volume 3, Issue 3 | March-April-2018 | http:// ijsrcseit.com 1298

 A. L. Choodarathnakara et al. Int J S Res CSE & IT. 2018 Mar-Apr;3(3) : 1296-1304

Figure 2. Flowchart of the Experimental Setup

1) Flowchart of the Proposed Methodology: The

Figure 2 shows the flowchart of the experimental
setup.

2) Gaussian Filtering: In this work, Gaussian filter is

Table II LANDSAT 7 ETM+ Bands used with Ground used to illustrate how frequency domain filters can
Features be used as guides for specifying the coefficients of
Ground feature Band used some of the small masks. Filters based on Gaussian
Water 1,2,3; 1,2,4; 1,4,5 forward and Inverse Fourier Transform of a Gaussian
Urban 1,2,3; 1,4,5
function is the real Gaussian functions.
Farmland 1,2,3; 1,4,5
Forest 1,2,3; 1,4,5
Salt scald 1,2,3
Scrub 1,4,5
Vegetation 1,4,7 Let, H(u) denote the 1-D frequency domain Gaussian
filter.
H(u)=A ……. (2)
C. Methodology
Where, is the standard deviation of the Gaussian
START curve. The corresponding filter in the spatial domain
obtained by taking the inverse Fourier transform of
RAW LANDSAT IMAGE H(u).
h(x)=√ A …..... (3)
1.1) Gaussian Low-pass Filters: The 2-D of
Gaussian low pass filter is given by,
GUASSIAN FILTER WITH VARYING
WINDOW SIZE AND S.D
H(u,v)= …… (4)
Where, D(u,v) is the distance from center of the
frequency rectangle. As before, is measure of
STATISTICAL ANALYSIS spread about the center. By letting , we can
express the filter using the notation of the other
filters
SELECTING PROPER WINDOW SIZE
WITH STD DEVIATION H(u,v)= ….. (5)
Where, is the cutoff frequency, when D (u, v) =
, the GLPF is down to 0.607 of its maximum value.
1.2) Gaussian High-pass Filters: The transfer
END function of the GHPF with cut-off frequency locus at
a distance from center of the frequency rectangle
is given by,

H(u,v)=1- ….. (6)

Volume 3, Issue 3 | March-April-2018 | http:// ijsrcseit.com 1299

 A. L. Choodarathnakara et al. Int J S Res CSE & IT. 2018 Mar-Apr;3(3) : 1296-1304

IV. RESULTS AND DISCUSSION The Figure 6 shows the input data from LANDSAT 7
ETM+ satellite image with 28.5m spatial resolution.
A. Matlab Window for Gaussian Filter The Figure 7 shows conversion to gray scale.
The Figure 3 shows the main interface for the
MATLAB and Figure 4 shows the code of Gaussian
filter written in MATLAB language and its execution
in Command Window.

Figure 6. Input data from Landsat (28.5 m)

Figure 3. The main interface for the MATLAB.

Figure 7. Conversion to gray scale

Figure 4. The code of Gaussian filter written in
MATLAB language.

Figure 5. Command window to run Gaussian Figure 8. Window 3X3 with Standard Deviation
filtering code 0.375

Volume 3, Issue 3 | March-April-2018 | http:// ijsrcseit.com 1300

 A. L. Choodarathnakara et al. Int J S Res CSE & IT. 2018 Mar-Apr;3(3) : 1296-1304

The Figures 8, 9, 10 & 11 shows the images obtained

for 3x3 bandwidth with different Standard Deviation
0.375, 0.75, 1.5 & 3 respectively.

Figure 12. Window 5X5 with Standard Deviation

0.375

Figure 9. Window 3X3 with Standard Deviation 0.75

Figure 13. Window 5X5 with Standard Deviation

0.75

Figure 10.Window 3X3 with Standard Deviation 1.5

Figure 14. Window 5X5 with Standard Deviation 1.5

Figure 11. Window 3X3 with Standard Deviation 3

Volume 3, Issue 3 | March-April-2018 | http:// ijsrcseit.com 1301

 A. L. Choodarathnakara et al. Int J S Res CSE & IT. 2018 Mar-Apr;3(3) : 1296-1304

Figure 15. Window 5X5 with Standard Deviation 3

Figure 18. Window 7X7 with Standard Deviation 1.5

Figure 16. Window 7X7 with Standard Deviation Figure 19. Window 7X7 with Standard Deviation
0.375 3

Figure 17.Window 7X7 with Standard Deviation 0.75 Figure 20. Window 9X9 with Standard Deviation
0.375

Volume 3, Issue 3 | March-April-2018 | http:// ijsrcseit.com 1302

 A. L. Choodarathnakara et al. Int J S Res CSE & IT. 2018 Mar-Apr;3(3) : 1296-1304

The Figures 20, 21, 22 & 23 shows the images

obtained for 9x9 bandwidth with different Standard
Deviation 0.375, 0.75, 1.5 & 3 respectively.
Table 3. Gaussian Filter Statistical Results for
different window sizes
Filter Filter
Standard
Window Window Min Max Mean SNR
Deviation
SD Size
Original
Image 0 255 146.928 67.333 2.182
Grayscale 0 255 146.928 67.333 2.182
Figure 21. Window 9X9 with Standard Deviation 3 3x3 0 255 146.819 57.748 2.542
3 5x5 0 255 145.707 52.121 2.795
0.75
3 7x7 0 255 144.704 52.486 2.757
3 9x9 0 255 144.202 51.889 2.779
1.5 3x3 0 255 146.823 58.005 2.531
1.5 5x5 0 255 145.715 55.318 2.634
1.5 7x7 0 255 144.733 55.037 2.629
1.5 9x9 0 255 144.186 55.811 2.583
0.75 3x3 0 255 146.835 59.503 2.468
0.75 5x5 0 255 145.748 59.645 2.444
0.75 7x7 0 255 144.736 60.621 2.388
0.75 9x9 0 255 144.196 61.566 2.342
0.375 3x3 0 255 146.535 66.903 2.190
0.375 5x5 0 255 145.419 67.640 2.149
0.375 7x7 0 255 144.408 68.462 2.109
Figure 22. Window 9X9 with Standard Deviation 1.5 0.375 9x9 0 255 143.859 69.238 2.077

V. CONCLUSION & SCOPE FOR FUTURE

WORK
A. Conclusions
1) Gaussian Filter: For satellite image with high
resolution around 30m window 5X5 is recommended.
It gives the best result and the best enhancement of
the image with preserving the edges.

2) Future Work: These filtering techniques can be

implanted further for different satellite data products.
Figure23. Window 9X9 with Standard Deviation 3
These filtering techniques can be implanted further
for more than 9x9 window sizes to identify the
The Figures 12, 13, 14 & 15 shows the images
impact quality bandwidth on satellite image quality.
obtained for 5x5 bandwidth with different Standard
Deviation 0.375, 0.75, 1.5 & 3 respectively.
VI. ACKNOWLEDGEMENT
The Figures 16, 17, 18 & 19 shows the images
Special thanks to Mr. Nasser Mustafa Saleh Abdin for
obtained for 7x7 bandwidth with different Standard
encouragement, guidance and other contributions to
Deviation 0.375, 0.75, 1.5 & 3 respectively.
this study. All the Landsat 7 ETM+ data used in this
work is downloaded from the USGS web
(http://glovis.usgs.gov), Special thanks to the group.

Volume 3, Issue 3 | March-April-2018 | http:// ijsrcseit.com 1303

 A. L. Choodarathnakara et al. Int J S Res CSE & IT. 2018 Mar-Apr;3(3) : 1296-1304

The authors would like to thank anonymous Pattern Anal Machine Intell. Vol. PAMI-8, Jan,
reviewers for their valuable comments. 1986, pp. 15–25.
[9]. Cui F. Y. Zou L. J. and Song B, “Edge Feature
VII. REFERENCES Extraction Based on Digital Image Processing
Techniques“, International Conference on
[1]. Hyder Ali, Sukanesh and Fellow, "An Edge Automation and Logistics, IEEE, 2008, pp.
Preserving Denoising Technique for MR 2320-2324.
Images using Curvelet Transform", [10]. Nadernejad E. Sharifzadeh S. and Hassanpour
Interdisciplinary Journal, Vol. 91, May 2010, H, “Edge Detection Techniques: Evaluations
pp.3-8. and Comparisons“, Applied Mathematical
[2]. Shujun Fu, Qiuqi Ruan, Wenqia Wang and Yu Sciences, Vol. 2, No. 31, 2008, pp.1507- 1520.
Li, "Adaptive Anisotropic Diffusion for [11]. Amit Chaudhary, Tarun Gulati, “Segmenting
Ultrasonic Image Denoising and Edge Digital Images Using Edge Detection“,
Enhancement", International Journal of International Journal of Application or
Information Technology, Vol. 2, No. 4, 2006, Innovation in Engineering & Management
pp. 284-287. (IJAIEM), Volume 2, Issue 5, May 2013.
[3]. Shujun Fu, QiuqiRuan, Wenqia Wang and Yu [12]. Simranjit Singh Walia, Gagandeep Singh,
Li, "Feature Preserving Nonlinear Diffusion for “Color based Edge detection techniques – A
Ultrasonic Image Denoising and Edge Review“, International Journal of Engineering
Enhancement", World Academy of Science, and Innovative Technology (IJEIT) Volume 3,
Engineering and Technology, Vol. 2, February Issue 9, March 2014.
2005, pp. 148-151. [13]. Er. Komal Sharma, Er. Navneet Kaur,
[4]. Tanaphol Thaipanich and Jay Kuo, "An “Comparative Analysis of Various Edge
Adaptive Nonloca Means Scheme for Medical Detection Techniques“, International Journal
Image Denoising", In Proceeding of SPIE of Advanced Research in Computer Science
Medical Imaging.Vol.7623,San Diego, CA, and Software Engineering, Vol. 3, Issue 12,
USA, February 2010. December 2013.
[5]. Su Cheol Kang and Seung Hong Hong, "A [14]. John R. Jensen and Dave C. Cowen,“ Remote
Speckle Reduction Filter using Wavelet- Sensing of Urban/Suburban Infrastructure and
Based Methods for Medical Imaging Socio-Economic Attributes“, Photogrammetric
Application", In Proceedings of 23rd Annual Engineering &Remote Sensing, 1999, pp. 611-
International Conference of the IEEE 622.
Engineering in Medicine and Biology Society,
Vol.3, pp. 2480-2483.
[6]. C. Chao and A. P. Dhawan, “Edge detection
using Hopfield neural network“, Proc. SPIE,
Vol. 2243, 242 (1994); doi:10.1117/12.169971.
[7]. D. Marr and E. Hildreth, “Theory of edge
detection“ Proc. Royal Society of London, B,
1980, 207, pp. 187–217.
[8]. A. L. Yuille and T. A. Poggio, “Scaling
theorems for zero-crossings“, IEEE Trans.

Volume 3, Issue 3 | March-April-2018 | http:// ijsrcseit.com 1304