Volume 4, Issue 6, June – 2019 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Image De-Noising using Method Noise Thresholding

Dr. H. N. Suresh Bhattramakki Sahana
Professor and PG Coordinator Student
Department of EIE, Signal Processing Department of EIE, Signal Processing
Bangalore Institute of Technology (BIT) Bangalore Institute of Technology (BIT)
Bengaluru, India Bengaluru, India

Abstract:- Synthetic aperture radar (SAR) images are

usually distorted by speckle noise. This problem is In the midst of image acquirement or transmission
resolved by de-speckling the noisy image. The proposed computerized images are debased by various disturbances
method denoises SAR image by applying discrete primarily known as impulse noise. During the time spent
wavelet transform (DWT). This scheme is proposed in imaging the noise taints the images by substitution of
db4 type wavelet transform. The difference between a couple of pixels of unique image with some new pixels,
noisy image and its denoised image, which is also these new pixels have luminance which can be inside the
defined as method noise is applied thresholding for both base luminance range or it tends to be past the luminance
approximate and detail part of wavelet decomposition. go named as motivation clamor. This happens because of
This article presents an improved and enhanced flawed exchanging. Amid transmission normally these sorts
filtering technique using median and Crimmins filter as of loud images happen.
pre-processing and post processing technique, for main
method DWT, to avoid blurring and to increase the The accomplishment of the applications relies upon
clarity of an image. The performance of the scheme is the expulsion of noise from the image information as the
evaluated by calculating Peak signal to noise ratio evacuation of the clamor majorly affects the execution of
(PSNR), Mean square error (MSE), Percentage of image handling. Noise removal is a troublesome procedure
Spoiled Pixels (PSP), structural similarity index as the noise reclamation channel must not distort the
measure (SSIM). The execution of the subsequent valuable information in the image. There are different
improved strategy is compared with some standard strategies which are proposed for motivation of noise
techniques and it indicates better outcomes regarding expulsion from advanced images.
numerical and visual examination.
SAR [Synthetic-aperture radar] is utilized to make 2D
Keywords:- SAR Images, Crimmins Filter, Median images or 3D images which are otherwise called landscapes.
Filter ,Method Noise Radar reception apparatus is utilized on an objective district
by SAR which gives better spatial resolution. SAR
I. INTRODUCTION movement is superior to anything conventional beam-
scanning radars. SAR is set on air ship, a moving stage and
As synthetic aperture radar (SAR) has been broadly it is started from SLAR [side looking airborne radar].SAR
utilized almost in each field, SAR image de-noising turned gadget separation to the objective is equal to the time taken
into a significant research field. Most of the inquiries about by the radar pulses. Immense synthetic antenna aperture is
image de-noising have been centered on how to utilize the made as the receiving wire measure. In light of the size the
compelling data of images to diminish noise. With the image resolution is chosen. Bigger gap size would give
improvement of sparse representation, the blend of adaptive great image resolution. It doesn't make a difference if the
dictionary and low rank model features its favorable receiving wire is extensive or in the event that it is a
circumstances. Existing de-noising calculations are not moving reception apparatus. In spite of the fact that the
impeccable. The distinction between an original image and physical status of reception apparatus is little it gives a
its de-noised image, which is additionally characterized as high-resolution image.
method noise contains image data. An ever-increasing
number of scientists have understood the significance of The 3D image preparing is achieved basically in 2
strategy disturbance and proposed many preparing steps:
techniques to improve the de-noising execution. For
instance, Chen et al. planned a sort of different separating  The azimuth and range bearing are engaged for
to manage the leftover image in [6]. generating 2D images.
 Next, a digital elevation model (DEM) is utilized to
The success of the applications depends on the calculate the phase difference between complex images
removal of noise from the image data as the removal of the which is resolved from various look angles to recoup
noise has a major impact on the performance of image the height information.
processing. Noise removal is a difficult process as the noise
restoration filter must not garble the useful info in the Then the height information, alongside azimuth-range
image. There are various methods which are proposed for provides third dimension.
impulse noise removal from digital images.

IJISRT19JU265 www.ijisrt.com 124

Volume 4, Issue 6, June – 2019 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
II. METHODOLOGY

The undermined image is gone through the Pre-

handling based median filter separating to beginning drive
noise clearing at that point. This pre-processed image is
passed through Wavelet decomposition up to level 2 and at
each level’s row and column stage, method noise
thresholding is applied. These wavelet coefficients obtained
are then used for reconstruction of an image. The resultant
image obtained after reconstruction is then passed through
crimmins filter for post-processing, which smoothens the
image and final image is obtained.

A. Noisy Image:
Noisy image is got by adding sound on the original
gray images of the given size 512×512. Here 512×512
designates that an image is having 512 rows and 512
columns. Consider salt and pepper noise and impulse noise.
Fig 1:- SAR Imaging
B. Pre-Processing Median Filter:
For SAR image creation, radio wave pulses are lit up Next step is where the noisy image is headed through
to target scene each pulse echo is recorded and obtained. the pre-processing median filter. Image filters are primarily
Pulse transmission and echoes receivable is finished used for removal of noise, detection of edge, sharpen
utilizing reception apparatus of single beam. The contrast. This removes artifacts from the original image.
wavelengths of a meter are decreased to a few millimeters. Median filter is a non-linear digital filtering method,
Movement of SAR gadget set on flying machine will do frequently utilized to evacuate noise from an image or
changes in target when there is change in time and area. signal. Such noise lessening is a pre-processing step to
Recorded radar echoes of numerous reception apparatus make strides the comes about of afterward handling (for
places of progressive radars are acquired and signal case, edge location on an image). Median filtering is
processing of the equivalent gives consolidating of various exceptionally broadly utilized in advanced image handling
radio wire positions. This procedure is synthetic antenna since, beneath certain conditions, it preserves edges
aperture which delivers high resolution images. whereas eradicates noise.

SAR image is delivered by exchange of C. Discrete Wavelet De-Noising:

electromagnetic waves that is space borne or it tends to be The discrete Wavelet De-noising-based image de-
conceived in air close to surface target and consistently noising has below steps. Apply 2D wavelet change and
process the returned signs from different targets. The change the noisy image into orthogonal area. Apply method
consistent preparing prompts dusty impact as a result of noise thresholding (here soft thresholding is used) to the
which the image becomes noisy, these granular noises will uproarious detail coefficients of the wavelet change.
diminish resolution of the image and frustrate elucidation Execute inverse discrete wavelet transform to get the de-
of image and investigation of image. noised image. Here, the edge assumes an imperative part in
the procedure used for de-noising. Finding an ideal limit is
The aim of SNR de-noising project is to reduce the a little difficult procedure. A little limit esteem will hold
impulse noise present in images by using algorithms along wavelet coefficients while a vast edge esteem prompts the
with filters, comparison Random Forest method and loss of wavelet coefficients that convey points of interest of
method noise thresholding, preserving textural contents and an image.
features of structure and accuracy is increased in detecting
the noisy pixels. D. Post Processing Using Crimmins Filter :
Crimmins filter is basically used for de-speckling and
The rest of the thesis follows as, literature survey is smoothening of an image. The calculation employments a
explained in second section explains, explanation of the non-linear noise diminishment technique which compares
methodology used, is given in third section, experimental the escalated of each pixel in an image with those of its 8
results and comparison of parameters are provided in fourth closest neighbors and, based upon the relative values,
section and fifth section is all about the conclusion and increases or decrements the esteem of the pixel in address
future scope. such that it gets to be more representative of its
environment.

IJISRT19JU265 www.ijisrt.com 125

Volume 4, Issue 6, June – 2019 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
E. Random Forest Classifier: III. EXPERIMENTAL RESULTS
Random forests is also known as random decision
forests. Random decision forests are ensemble A set of parameters are used to calculate the results of
learning method which is used for classification. It is also the reconstructed images. Different parameters define the
used for regression. Random forests operation is carried out performance of proposed algorithm. The results obtained
by constructing decision trees multitude during training and with the proposed methods have been compared with the
the output is mode of classes which is used in classification existing methodologies.
or mean prediction which is used in part of regression.
A. Peak Signal to Noise Ratio
A supervised learning algorithm termed as Random The peak signal-to-noise ratio figures, in decibels,
Forest. This method as the name indicates a forest is created between two images. This ratio is frequently utilized as a
and it is made random. The forest built is ensemble of quality measurement. It is used between the original and a
Decision Trees. These Decision Trees are trained by compacted image. The higher the PSNR, the quality of
“bagging” method. Bagging method is a learning model compacted image is better.
where the overall result is increased.
B. Mean Square Error [MSE]
Random forest advantages are, it is applied for The Mean Square Error is one of the metric and
classification problems, applied for regression problems and the Peak Signal to Noise Ratio (PSNR) is another error
it used majorly in today’s machine learning systems. metrics. They are used to do comparison of image
Steps for de-noising SAR image is as follows: compression quality. The MSE signifies the cumulative
squared error. It occurs between the compressed image and
 Step 1: Input the original noise free Synthetic aperture the original image. PSNR signifies a measure of peak error.
radar (SAR) image to the summation block. The error will be reduced as the MSE value is reduced.
 Step 2: The second input is impulse also called as salt
and pepper noise of anticipated noise density. C. Structural Similarity Index Measure [SSIM]
 Step 3: The output of summation is now the SAR image Structural similarity [SSIM] index method computes
is noisy and is corrupted with impulse noise of certain the similarity between two images. The measure of two
noise density. windows which has the same size.
 Step 4: SAR image corrupted by this impulse noise is
now provided to the median filter for preprocessing of D. Percentage of Spoiled Pixels
the image. Here around 20% to 30% of the noise It is the measure which gives unaffected original
present in the image is removed. pixels and swapped with different gray value after filtering.
 Step 5: On this pre-processed image, DWT (discrete Small value of PSP will give improved performance. High
wavelet transforms) is applied, where detail and value of POSP removes original image property.
approximate wavelet coefficients are obtained and these
are subjected to method noise thresholding. Image 1a: RF SAR image with Noise density 0.4
 Step 6: Now from this decomposition, we obtain
wavelet coefficients from which we then reconstruct an
image.
 Step 7: In final stage Crimmins filter is applied on to the
image to further reduce the amount of noise in the
image and smoothen the image.

Mat lab is a well-organized programming language for

matrix processing. This programming language can be (A) (B)
termed as Matrix Laboratory. It contains readymade
functions for manipulations and visualization of images. It
allows program as a modular structure. Mat lab
[matrix laboratory] is known as multi-
paradigm computation environment and it is also known as
fourth-generation language used for programming.
An exclusive language technologically advanced through
Mathworks. (C) (D)
Fig 4:- (A) Depicts the Original SAR Image, (B) Depicts
the Image after N Oise is Added, (C) Depicts the Image
after Pre-Processing, (D) Depicts the Filtered SAR Image.

IJISRT19JU265 www.ijisrt.com 126

Volume 4, Issue 6, June – 2019 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

Table 1:- Comparison of RF and Method Noise Thresholding for 0.4 Noise Density on Standard SAR Image

IV. CONCLUSION AND FUTURE SCOPE REFERENCES

This project proposes a novel and productive method [1]. “A new SAR image despeckling using directional
for de-noising the synthetic aperture radar images. This smoothing filter and method noise thresholding “by
implementation is done in 2 stages. First the pre-processing Prabhishek Singh , Raj Shree published in
stage is implemented, in which the image is resized and Engineering Science and Technology, an International
RGB to gray conversion is performed in presence of Journal, 14 May 2018 , pp 180-189
median filter (2 dimension). Median filter is used to reduce [2]. “Weighted Nuclear Norm Minimization with
the impulse noise content to certain level. Then the pre- Application to Image Denoising” by Shuhang Gu, Lei
processed image is again filtered using the DB4 wavelet Zhang, Wangmeng Zuo, Xiangchu Feng published in
transform de-noisier. In this proposed scheme of filtration, IEEE Conference on Computer Vision and Pattern
there is a wavelet based filtering, hence the computational Recognition , October 2014, pp 155-162
complexity increases but improved performance has been [3]. “Image denoising using local Wiener filter and its
achieved in restoring results as compared to the previously method noise” by Xiaobo Zhang published in elsevier
reported schemes. There could be chances that this pre- magazine , September 2017, pp 6821-6828
filtered image can still be encountered with malicious [4]. ” Image Denoising using Wavelet Transform
pixels hence finally post-processing is done using crimmins Method” by Vikas Gupta, Rajesh Mahle, Raviprakash
filter. Crimmins filter is used to smoothen the reconstructed S Shriwas, September 2012, pp 258-267.
image and also removes the noise. The qualitative [5]. "Translation-invariant denoising using
investigation is done by analyzing the visual appearance of multiwavelets" by T. D. Bui and G. Y. Chen
the despeckled image counting the parameters just like the published in IEEE Transactions on Signal Processing,
presence of the artifacts; edge conservation; visibility of Vol.46, No.12, pp.3414-3420, 1998.
low differentiate objects; surface conservation and [6]. “Noise reduction in SAR images by using fusion of
conservation of homogeneous and non-homogeneous stationary and discrete wavelet transforms” by R.
locales. The quantitative examination is done utilizing Sunil, CHPulla Rao and A. Sarada, published in
parameters like PSNR, MSE, SSIM and PSP .These International Journal of Science, Engineering and
parameters shift with distinctive SAR images. The Technology Research , Volume 5, Issue 2, February
statistical results of the proposed strategy are normalized 2016 , pp515-521.
and empowering, which can be advance improved by the [7]. “Efficient techniques for de-noising of highly
intelligent utilize of method noise. corrupted impulse noise images” by Suresh Velaga,
Sridhar published in International Journal of Soft
Various methods to denoise a satellite image has been Computing and Engineering ,Volume 2, Issue 4,
learnt as part of this project. Analyzing an image and the September 2012, pp 255-258.
best suitable method to be selected, which would make the [8]. “High Density Impulse Noise Removal Utilizing
image appear more attractive for most of the images has Switching Bilateral Filter”by T. Veerakumar, S.
been learnt. For example in satellite images, Crimmins Esakkirajan and IlaVennila, published in International
filters proves to be best smoothening filter compared to lee journal of circuits, systems and signal processing,
filter, DSF (directional smoothing filter),average filter etc. 2012, Issue 3, Volume 6, pp 189-196.
All images cannot be de-noised with same efficiency with [9]. “Gaussian Noise Removal by a New Approach of
one single method as the parameters keep changing for Discrete Wavelet Transform”, by PoojaPandey,
different images. Hence the coding skill in MATLAB and PraveenChouksey, published in International Journal
study of various methodologies has been improved as part of Science, Engineering and Technology Research
of this project. ,Volume 5, Issue 2, February 2016.
[10]. “A Novel Filtering Scheme for Impulse Noise
Cancellation”,by G Panda, BMajhi , P K Dash and S
K Meher published in IETE journal of research,
March 2015, volume 48, pp 163-170.

IJISRT19JU265 www.ijisrt.com 127

Volume 4, Issue 6, June – 2019 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
[11]. “Review Of Impulse Noise Reduction Techniques”,
by Manohar AnnappaKoli, published in International
Journal On Computer Science And Engineering ,
Volume 4, February 2012, pp 184-196.
[12]. “Image De-noising using Median Filter and DWT
Adaptive Wavelet Threshold”, by
M.G.Srinivasa,Dhanushree.V published in IOSR
Journal of VLSI and Signal Processing, Volume 5,
Issue 3, pp 50-57.
[13]. “Classification of Single and Multiple PQ
Disturbances Based on DWT and RF Classifiers”, by
Anton Domini Sta. Cruz, published in Procurement of
the 2017 IEEE , November 8, 2017, pp 2182-2187.
[14]. “Detection and Removal of Salt and Pepper noise in
images by using Improved Median Filter”,by
S.Deivalakshmi, P.Palanisamy and S.Sarath, IEEE,
2011, pp 363-368.
[15]. “Adaptive Image De-noising Using Wavelet
Thresholding”, by Liwen Dong, oublished in Third
International Conference on Information Science and
Technology, March 2013.

IJISRT19JU265 www.ijisrt.com 128