Download as pdf or txt
Download as pdf or txt
You are on page 1of 4

2019 19th International Symposium on Communications and Information Technologies

(ISCIT)

Performance Inspection of Denoising Method


Implemented by Decision Based Adaptive Median
Filter for Using under RVIN Condition
Vorapoj Patanavijit Kornkamol Thakulsukanant
Assumption University of Thailand Assumption University of Thailand
Bangkok, Thailand Bangkok, Thailand
Email: Patanavijit@yahoo.com Email: kthakulsukanant@yahoo.com

Abstract—In 2010, the DBAMF (decision based adaptive implemented by DBAMF for RVIN (Random Value
median filter) denoising method based on ROAD that was Implosive Noise) condition.
proposed for eradicating an impulsive noise because ROAD
(Rank Order Absolute Difference) has the obviously high II. DISCIPLINED IDEA OF DBAMF DENOSING METHOD
performance in noisy classification and is mathematically In this thesis article, the denigrated image is
developed from similar measurement method in computer
mathematically prescribed as Y and the intensity amplitude
vision for identifying whether the pixel is contaminated by
impulsive noise or the noise-free. For that reason, the scholar
at address ( i , j ) is mathematically prescribed y ( i, j ) . This
thesis article intents to inspect the performance of the denosing DBAMF method comprises of noise classification sub-
method implemented by DBAMF for RVIN (Random Value procedure (using ROAD or Rank Order Absolute
Implosive Noise) condition. For inspecting the performance Difference), which is extravagantly explicated in section II-
and its impediment of the DBAMF denoising method, the two A, and enhanced sub-procedure (using SMF [2,6]) for
worldwide assessment images integrated of Girl and Airplane denoising random-value impulse noise at some power or
has been manipulated in these performance inspections under density), which is extravagantly explicated in section II-B.
RVIN condition by originally denigrated by accumulating the
RIVN at generous power. Later, every denigrated image is
enhanced by the denosing method implemented by DBAMF.
For indicating the DBAMF performance, the classical SMF A. Noise Classification Sub-Procedure
(Standard Median Filter) and AMF (Adaptive Median Filter) The data processing idea of the noise classification sub-
are comparatively integrated in this performance inspection. procedure can be extravagantly explicated as following.

Keywords— DBAMF (Decision Based Adaptive Median 1. The 3 × 3 ( w = 3 ) square region of image, where the
Filter), RVIN (Random Value Implosive Noise), Image middle point is located at ( i , j ) , is mathematically
Denosing, AMF (Adaptive Median Filter prescribed as W3×3 .
I. RELEVANT INSPECTED RESEARCH
2. For each pixel, the normalized absolute difference
In general, impulsive noise, which are distributed into ( Ds ,t ( yi , j ) ) between all pixels in square region and the
two categories: Salt and Pepper noise and random-value
impulse noise from the mathematical designing aspect, is midpoint at ( i, j ) location are mathematically
generated by generous argumentations for instant CCD prescribed using the following algebraic term.
deficiency, inaccuracy of analog-to-digital conversion, Ds ,t ( yi , j ) = yi , j − y s ,t 255 (1)
communication indiscretion or memory mislocating , etc.
Afterward, denoising method has been inspected for more 3. The group of all normalized absolute difference
than three decades in consequent of the specification of Ds ,t ( y i , j ) are arranged and stored only for the five
advance digital image processing applications: SR (super smallest values and, next, the first moment of statistics
resolution), facial emotion classification, eye of these five smallest values is prescribed, so called
acknowledgment, etc. At initial, the SMF (Standard Median
normalized ROAD at the ( i , j ) pixel location. The first
Filter) [3,7] was desired for solely phasing out SPN (Salt and
Pepper Noise) and has become one of the most worldwide moment normalized ROAD value can be
SPN denoising method for the reason of great efficient and mathematically prescribed using the following algebraic
non-complication. After fifteen years, generous desired term.
denoising methods [1,2,4,5,8], which are developed from 5

SMF, have been desired for phasing out SPN. For that ROAD =1m =5 D (y )
5  s ,t i, j
(2)
m =1
reason, the decision based adaptive median filter (DBAMF)
4. For each pixel, the first moment normalized ROAD
[6], which are developed from SMF, is desired in 2010. This
value of the interested pixel is correlated with a steady
DBAMF method comprises of noise classification sub-
number T0 , which is adjusted between 0.00 and 1.00 for
procedure (using ROAD or Rank Order Absolute Difference)
and enhanced sub-procedure (using SMF [2,6]) for denoising all images from simulated highest PSNR [6], where the
random-value impulse noise at some power or density. NDM (Noise Detected Matrix) can be mathematically
Unfortunately, there have no inspecting of the performance prescribed using the following algebraic term.
and its impediment of the DBAMF denoising method when
this denoising method is implemented under the RIVN at 1 , ROAD m =5 ≥ T0 (3)
NDM = 
generous power. Thereby this scholarly thesis article intents 0 , ROAD m =5 < T0
to inspect the performance of the denosing method

XXX-X-XXXX-XXXX-X/XX/$XX.00 ©20XX IEEE

978-1-7281-5009-3/19/$31.00 ©2019 IEEE 422


2019 19th International Symposium on Communications and Information Technologies
(ISCIT)

From the above noise classification sub-procedure of the B. Enhanced Sub-Procedure


DBAMF, we can be extravagantly explicated as the This enhanced sub-procedure is enhanced only the
figure 1. denigrated pixels, which are defined by previous sub-
procedure, and the sub-procedure of the DBAMF can be
Start extravagantly explicated as following procedure.
1. The 3 × 3 ( w = 3 ) square region of NDM (noise detected
matrix), where the middle point is located at ( i , j ) , is
Read Noisy Image Y
mathematically prescribed as W3×3 .

j =1 2. For each interested square region, if the number of


noiseless pixels in this square region is less than three
pixels then the size of this square region WN × N is
expanded by 1 and repeated Step 2 again.
No 3. For each interested square region, if the number of
j ≤ y-Dimension noiseless pixels in this square region is equal to or more
than three pixels then the denigrated pixels are enhanced
Yes by using the following algebraic term.
i =1
{
Yˆi , j = median Yi − s, j −t ( s, t ) ∈WF } (4)

4. The identical procedure is reprocessed for all denigrated


No pixel, which are defined by previous sub-procedure, in
i ≤ x-Dimension this image.
Yes III. EXPERIMENTAL PERFORMANCE INSPECTION OF DBAMF
DENOISING METHOD
Set W3×3 with center at y ( i, j ) In this experimental inspection, the simulated program is
the MATLAB, which is run on workstation with the
hardware condition: Core i7-6700HQ at 16 GB Memory.
Ds ,t ( yi , j ) = yi , j − ys ,t
Each workstation run on these generous official images (that
are comprised of Girl and Airplane), which are denigrated
5
ROAD m =5 =  Ds ,t ( yi , j )
by accumulating the RIVN at generous power. All
m =1 denigrated images are manipulated for reaching the upper
Yes
bound of PSNR of DBAMF denoising method by running
the DBAMF noise classification sub-procedure (for
( Noisy ) classifying noisy pixels from noise-free pixels) and the
ROAD m=5 ≥ 40
DBAMF enhancement sub-procedure (for enhancing noisy
No ( Noiseless ) pixels to be noise-free pixels).
y ( i, j ) is a noiseless pixel A. The Experimental Inspection of SteadyNumber T0
Noise_Detection ( i, j ) = 0
The performance of the DBAMF noise classification sub-
procedure is forcefully counted on a steady number T0 ,
which is adjusted between 0.00 and 1.00. Accordingly, this
experiment comprehensively inspects the best steady number
y ( i, j ) is a noisy pixel
T0 , for reaching the upper bound of PSNR of DBAMF
Noise_Detection ( i, j ) = 1
denoising method. Generous digital images (that are
comprised of Girl and Airplane) are manipulated to inspect
the steady number T0 by adjusted between 0.00 and 0.50 at
i = i +1
0.025 incremented step as extravagantly explicating in the
Table 1 to Table 2, respectively.
From these experimental comparative inspection of Girl,
j = j +1 the best steady number T0 should be 0.0792 ± 0.0231 or be
regulated from 0.075 to 0.100 for reaching the upper bound
of PSNR of DBAMF denoising method.
End
From these experimental comparative inspection of
Airplane, the best steady number T0 should be 0.0847 ±
Figure 1. Disciplined idea of noise classification sub-procedure of the
0.0311 or be regulated from 0.075 to 0.100 for reaching the
DBAMF upper bound of PSNR of DBAMF denoising method.

423
2019 19th International Symposium on Communications and Information Technologies
(ISCIT)

B. The Experimental Inspection of Denoising Performance DBAMF for RVIN (Random Value Implosive Noise)
This sub-precedure intends to comparatively inspect the condition. In this experimental inspection, the best steady
denoising performance of DBAMF, AMF, SMF and Mean number T0 at each RIVN density, for reaching the upper
Filter of Girl and Airplane images as extravagantly bound of PSNR of DBAMF denoising method, is
explicating in the Table 3. comprehensively determined from this simulation on both
Girl and Airplane under the RIVN at generous power. Later,
For Girl image, the performance of the DBAMF
the experimental comparative inspection of DBAMF
denoising method is higher than SMF about 11.7268 ± denoising method and other well-known denoising methods
1.9080 dB, is higher than Mean Filter about 12.1264 ± such as SMF (Standard Median Filter), Mean Filter and
2.6805 dB and is higher than AMF about 10.5896 ± 1.3931 AMF (Adaptive Median Filter) are comprehensively
dB. determined. From this experimental comparative inspections
For Airplane image, the performance of the DBAMF and statistical analysis, the DBAMF denoising method has
denoising method is higher than SMF about 8.1136 ± 1.2859 dramatically better performance than other well-known
dB, is higher than Mean Filter about 9.2415 ± 1.8955 dB and denoising methods about 10-12 dB for Girl image and 7-9
dB for Airplane.
is higher than AMF about 7.9553 ± 1.4273 dB.
ACKNOWLEDGMENT
IV. EXPERIMENTAL SUMMARY
The research project was funded by Assumption
This scholar thesis article intents to inspect the
University.
performance of the denosing method implemented by

TABLE I. EXPERIMENTAL COMPARATIVE INSPECTION OF DENOISING PERFORMANCE IN PSNR AND STEADY NUMBER OF GIRL IMAGE (UNDER RVIN)

Fix PSNR (dB)


Constant
5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90
0.025 27.48 26.51 26.50 25.76 25.27 23.69 23.20 22.85 22.75 21.50 21.64 21.00 19.87 18.84 17.19 15.40 14.04 12.63
0.050 32.37 31.04 30.18 28.31 27.58 26.46 24.80 24.29 23.42 22.72 22.48 22.46 21.80 20.75 19.14 17.08 15.55 14.25
0.075 35.88 34.34 32.54 30.59 29.12 27.57 25.60 25.24 24.68 23.38 23.04 22.25 20.83 20.34 19.82 17.05 16.90 15.16
0.100 38.71 35.72 34.22 31.33 28.71 27.00 25.76 25.01 23.99 22.80 21.84 20.47 19.42 18.44 17.65 16.01 14.93 13.29
0.125 38.81 36.10 33.98 30.46 28.98 25.93 25.34 23.56 23.04 21.38 20.17 18.88 17.74 16.44 15.58 14.00 13.36 11.62
0.150 38.00 34.98 32.58 30.63 27.85 24.65 23.83 22.74 21.18 19.90 18.57 17.28 16.20 14.97 13.99 12.55 11.95 10.25
0.175 36.76 33.58 31.14 29.16 26.92 23.35 22.94 21.32 20.05 18.55 17.21 15.93 14.69 13.63 12.79 11.71 10.78 9.46
0.200 35.79 32.57 29.79 27.78 25.42 23.59 21.62 20.12 18.89 17.25 16.14 14.84 13.67 12.70 11.97 10.82 9.86 9.34
0.225 34.49 31.47 28.50 26.47 24.13 22.62 20.50 18.85 17.65 16.21 15.10 13.88 12.83 11.98 11.18 10.03 9.10 8.88
0.250 33.32 30.28 27.28 25.22 23.00 21.41 19.36 17.88 16.65 15.34 14.20 13.07 12.02 11.27 10.49 9.70 9.00 8.42
0.275 32.35 29.35 25.99 23.99 22.06 20.35 18.30 17.02 15.79 14.56 13.43 12.36 11.44 10.77 9.94 9.26 8.69 8.10
0.300 31.47 28.14 24.99 23.02 20.99 19.43 17.42 16.24 15.02 13.81 12.81 11.81 10.98 10.35 9.58 8.97 8.42 7.90
0.325 30.73 27.22 24.04 22.17 20.11 18.58 16.72 15.54 14.39 13.27 12.31 11.38 10.58 10.00 9.30 8.73 8.21 7.72
0.350 29.98 26.45 23.26 21.25 19.31 17.80 16.03 14.92 13.79 12.76 11.91 11.00 10.25 9.66 9.05 8.51 8.03 7.59
0.375 29.23 25.61 22.45 20.44 18.57 17.10 15.45 14.43 13.32 12.35 11.55 10.68 10.00 9.42 8.83 8.33 7.88 7.46
0.400 28.56 24.74 21.72 19.77 17.96 16.49 14.92 13.93 12.86 11.97 11.20 10.37 9.73 9.21 8.66 8.18 7.75 7.35
0.425 27.82 23.97 21.05 19.13 17.33 15.95 14.43 13.49 12.45 11.60 10.90 10.12 9.50 9.03 8.49 8.05 7.64 7.26
0.450 27.18 23.32 20.47 18.56 16.73 15.50 14.03 13.13 12.14 11.31 10.65 9.89 9.33 8.86 8.36 7.93 7.54 7.18
0.475 26.48 22.73 19.88 18.01 16.25 15.04 13.63 12.75 11.85 11.05 10.39 9.71 9.17 8.71 8.25 7.84 7.46 7.12
0.500 25.85 22.16 19.34 17.55 15.83 14.66 13.30 12.45 11.61 10.83 10.19 9.54 9.03 8.61 8.15 7.76 7.40 7.06

TABLE II. EXPERIMENTAL COMPARATIVE INSPECTION OF DENOISING PERFORMANCE IN PSNR AND STEADY NUMBER OF AIRPLANE IMAGE (UNDER RVIN)

Fix PSNR (dB)


Constant
5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90
0.025 26.16 25.88 24.85 23.74 23.58 22.43 22.70 22.05 21.69 21.42 20.40 20.17 19.36 18.63 17.30 15.77 14.01 12.62
0.050 28.67 28.07 27.67 25.83 25.45 24.58 24.49 23.34 22.91 22.23 22.77 21.51 21.17 20.16 19.45 17.96 15.93 13.25
0.075 31.65 30.20 29.69 28.02 26.25 25.79 25.03 24.04 23.52 23.05 22.14 21.34 20.44 19.75 18.90 17.68 16.49 14.75
0.100 34.19 32.77 30.99 28.12 26.39 26.46 25.18 24.06 22.63 22.08 20.99 20.08 19.31 18.35 17.36 15.33 14.14 12.15
0.125 35.36 32.88 31.28 29.75 27.15 25.13 24.10 23.04 21.58 20.98 19.59 18.56 17.78 16.70 15.51 13.97 13.19 12.19
0.150 35.03 32.40 30.36 28.58 27.13 24.84 23.23 21.75 20.38 19.48 18.34 17.26 16.27 15.25 14.18 13.20 11.91 10.87
0.175 34.22 31.13 29.15 27.31 25.75 23.94 21.95 20.52 19.08 18.33 17.03 15.97 15.00 13.90 13.18 12.23 11.30 10.51
0.200 33.32 30.06 28.01 26.11 24.50 22.79 19.13 19.73 18.24 17.27 15.96 15.01 13.90 13.24 12.30 11.57 10.78 10.25
0.225 32.31 29.11 26.82 24.90 23.26 21.62 20.10 18.59 17.24 16.26 15.17 14.23 13.31 12.52 11.43 10.94 10.27 9.75
0.250 31.41 28.15 25.67 23.85 22.23 20.54 19.18 17.83 16.42 15.56 14.47 13.54 12.68 11.92 11.25 10.43 9.98 9.41
0.275 30.54 27.04 24.68 22.88 21.34 19.58 18.29 16.96 15.71 14.86 13.85 12.97 12.19 11.48 10.87 10.19 9.73 9.21
0.300 29.80 26.18 23.82 22.05 20.39 18.77 17.51 16.23 15.06 14.26 13.31 12.51 11.76 11.12 10.58 9.94 9.51 9.04
0.325 28.93 25.44 23.08 21.27 19.63 18.04 16.85 15.63 14.52 13.76 12.90 12.12 11.43 10.85 10.32 9.74 9.35 8.90
0.350 28.30 24.79 22.35 20.51 18.93 17.36 16.20 15.06 14.03 13.31 12.54 11.78 11.16 10.60 10.12 9.57 9.21 8.78
0.375 27.59 24.00 21.73 19.83 18.27 16.79 15.67 14.59 13.65 12.96 12.22 11.52 10.92 10.39 9.95 9.43 9.10 8.70
0.400 26.93 23.36 21.10 19.24 17.72 16.27 15.21 14.18 13.26 12.63 11.95 11.25 10.72 10.21 9.81 9.31 9.00 8.63
0.425 26.38 22.81 20.53 18.72 17.21 15.86 14.82 13.79 12.97 12.36 11.70 11.04 10.56 10.07 9.68 9.21 8.93 8.57
0.450 25.81 22.31 20.03 18.25 16.79 15.47 14.47 13.52 12.70 12.12 11.51 10.88 10.42 9.95 9.58 9.14 8.87 8.53
0.475 25.35 21.83 19.57 17.83 16.39 15.14 14.17 13.26 12.46 11.90 11.33 10.74 10.30 9.85 9.50 9.08 8.81 8.49
0.500 24.86 21.40 19.13 17.44 16.05 14.84 13.90 13.05 12.26 11.74 11.18 10.62 10.20 9.77 9.44 9.03 8.78 8.47

424
2019 19th International Symposium on Communications and Information Technologies
(ISCIT)

REFERENCES
[1] H. Hwang and R. A. Haddad, Adaptive Median Filters New
Algorithms and Results, IEEE Trans. of Image Processing, 1994.
[2] R. H. Chan, C-W Ho and M. Nikilova, Salt&pepper Noise Removal
by Median-Type Noise Detectors and Detail-Preserving
Regularization, IEEE Trans. of IP, Vol. 14, No. 10, 2005.
[3] R. C. Gonzalez and R. E. Woods, Digital Image Processing, Prentice- (a-1)
Hall,Upper Saddle River,NJ, USA, 2nd edition, 2002. Original Image
[4] V. Patanavijit, The Bilateral Denoising Performance Influence of
Window, Spatial and Radiometric Variance, ICAICTA2015, 2015.
[5] Vorapoj Patanavijit, Performance Analysis of Denoising Algorithm
Based on Adaptive Median Filter Under Unsystematic Intensity
Impulse and Salt&Pepper Noise, The 6th International Electrical
Engineering Congress (iEECON2017), Krabi, Thailand, March 2018. (a-2) (D=10%) (a-3)
(IEEE Xplore) Corrupted Image 3 × 3 Median Filter
[6] V. R. Vijaykumar and P. Jothibasu, Decision Based Adatpvive
(PSNR=16.4414dB) (PSNR=21.9262dB)
Median Filter to Remove Blotches, Scratches, Streaks, Stripes and
Impulse Noise In Images, Proceedings of 2010 IEEE 17th
International Conference on Image Processing, September 2010.
[7] W. K. Pratt, “Median filtering,” Tech. Rep., Image Proc. Inst., Univ.
Southern California, Los Angeles, Sep. 1975.
[8] Yiqiu Dong, Raymond H. Chan, and Shufang Xu, A Detection
Statistic for Random-Valued Impulse Noise, IEEE Trans. on IP, April
2007.

TABLE I. THE EXPERIMENTAL RESULT OF COMPARATIVE


INSPECTION OF DENOISING METHODS

SPN PSNR (dB)


Noise Suppressing Technique
Operated Noise LR
Images Density Image Median Mean DBA (a-4) (a-5) (a-6)
AMF
(3x3) (3x3) MF 3 × 3 Mean Filter AMF Filter DBAMF
D=0.05 19.4746 25.7294 22.9512 30.0645 38.8057
(PSNR=19.8441dB)(PSNR=25.1361dB) (PSNR=36.1019dB)
D=0.10 16.4414 21.9262 19.8441 25.1361 36.1019
D=0.15 14.7453 19.5950 17.9764 21.8823 34.2195
D=0.20 13.4343 17.6088 16.5271 19.5726 31.3303
D=0.25 12.3889 16.1560 15.3585 17.6875 29.1155
D=0.30 11.6674 15.0930 14.5072 16.4551 27.5654
D=0.35 10.9450 13.9786 13.6076 14.9798 25.7597
D=0.40 10.3946 13.2355 12.9416 14.1249 25.2397
(b-1)
Girl D=0.45 9.8828 12.4743 12.3169 13.2217 24.6791
(256x256) Original Image
D=0.50 9.4483 11.7512 11.7438 12.3872 23.3770
D=0.55 9.0317 11.1155 11.2138 11.6862 23.0360
D=0.60 8.6223 10.4934 10.6637 10.9357 22.4646
D=0.65 8.2779 9.9552 10.2045 10.3392 21.7959
D=0.70 7.9734 9.4750 9.8004 9.8160 20.7477
D=0.75 7.6747 9.0237 9.3854 9.2725 19.8179
(b-2) (D=20%) (b-3)
D=0.80 7.3939 8.5547 8.9936 8.7676 17.0794 Corrupted Image 3 × 3 Median Filter
D=0.85 7.1268 8.1714 8.6289 8.3437 16.9034 (PSNR=13.4343dB) (PSNR=17.6088dB)
D=0.90 6.8638 7.7810 8.2609 7.9151 15.1617
D=0.05 20.9736 29.8334 24.3770 30.8826 35.3644
D=0.10 17.9926 25.8599 21.3810 26.4177 32.8789
D=0.15 16.2351 22.7263 19.5409 23.1149 31.2842
D=0.20 14.9461 20.6851 18.1387 20.9798 29.7503
D=0.25 14.0418 19.1468 17.1378 19.3355 27.1526
D=0.30 13.2327 17.6557 16.1865 17.7748 26.4568
D=0.35 12.5238 16.5218 15.3779 16.6101 25.1832
D=0.40 11.9817 15.5397 14.7203 15.5969 24.0605

Airplane D=0.45 11.4113 14.5808 14.0290 14.6121 23.5204


(256x256) D=0.50 11.0346 13.9361 13.5686 13.9691 23.0465
D=0.55 10.6198 13.2722 13.0438 13.2896 22.7662
D=0.60 10.1911 12.5664 12.5212 12.5764 21.5126
(b-4) (b-5) (b-6)
D=0.65 9.8753 12.0054 12.1049 12.0147 21.1709 3 × 3 Mean Filter AMF Filter DBAMF
D=0.70 9.5278 11.4369 11.6650 11.4404 20.1601 (PSNR=16.5271dB) (PSNR=19.5726dB) (PSNR=31.3303dB)
D=0.75 9.2632 10.9792 11.2880 10.9808 19.4473
D=0.80 8.9038 10.4369 10.8311 10.4369 17.9580 Fig. 2 Experimental comparative result of denoising method of the
D=0.85 8.6867 10.0780 10.5300 10.0781 16.4882
DBAMF
D=0.90 8.4097 9.6464 10.1623 9.6465 14.7503

425

You might also like