You are on page 1of 8

Journal of Ambient Intelligence and Humanized Computing

https://doi.org/10.1007/s12652-020-01737-1

ORIGINAL RESEARCH

Removal of high density salt and pepper noise in color image


through modified cascaded filter
B. Karthik1   · T. Krishna Kumar2 · S. P. Vijayaragavan3 · M. Sriram4

Received: 22 August 2019 / Accepted: 27 January 2020


© Springer-Verlag GmbH Germany, part of Springer Nature 2020

Abstract
A modified cascaded filter (MCF) for the restoration of color pictures that are extremely corrupted by salt and pepper noise
and random valued impulse noise are projected in this article. MCF algorithm restores the noisy pixel by trimmed median
value while other pixel values, 0’s and 255’s are present in the selected window using decision based median filter (DMF)
and when the pixel values are 0’s and 255’s then the noise pixel is replaced by mean value of all the elements present in the
selected window using unsymmetrical trimmed mean filtering. This modified cascaded filter proves better results than the
standard median filter, DMF, and alpha trimmed median filter, UTMF. The MCF is analyzed against various color images
and it provides superior peak signal-to-noise ratio and image enhancement factor.

Keywords  Color image · MCF · DMF · UTMF · PSNR · IEF

1 Introduction working out. In a similar fashion, the procedure of receiving


and analyzing visible expertise through digital computer is
Human beings are principally visual creatures, who rely solely named as digital image processing.
on the sense of imaginative and prescient. Imaginative and
prescient allows for humans to perceive and realize the arena
surrounding them in a greater method. Processing visible 2 Filtering techniques
know-how with the aid of the laptop has been drawing tremen-
dous concentration of investigators above the few epochs. The 2.1 Median filtering
procedure of getting and evaluating visual expertise by using
the humanoid species is mentioned to as vision, awareness or The median filter is the one authorized nonlinear filter used
for eliminating impulse noise, for the reason that the excel-
* B. Karthik lent de-noising energy and computational efficiency. Median
karthikguru33@gmail.com; karthik.ece@bharathuniv.ac.in Filter is a picture filter that is extra strong in instances where
T. Krishna Kumar white spots and black spots appear on the picture. For this
drkk@bharathuniv.ac.in procedure the middle price of the M × N window is considered
S. P. Vijayaragavan to exchange the black and white pixels (Eng and Ma 2001).
vijayaragavan.eee@bharathuniv.ac.in
M. Sriram 2.2 Arithmetic imply filtering
sriramm.cse@bharathuniv.ac.in
This can be a simple and straightforward mean filtering man-
1
Department of ECE, Bharath Institute of Higher Education ner. Let G be the given photograph and pixel x be a noisy
and Research, Chennai, India
pixel. The noisy pixel should have the worth 0[black] or
2
Bharath Institute of Higher Education and Research, 255[white]. Taking a moving kernel window of measure-
Chennai, India
ment m × n headquartered the pixel x, compute the normal
3
Department of EEE, Bharath Institute of Higher Education price of the noising snapshot G and replace the noising
and Research, Chennai, India
pixel worth through the ordinary worth. Sx,y be a set of
4
Department of CSE, Bharath Institute of Higher Education
and Research, Chennai, India

13
Vol.:(0123456789)
B. Karthik et al.

coordinates in a relocating kernel window (Oten and de does now not have any impulse, the UTMF desired now not
Figueiredo 1995). trim any ethics. On the opposite, the ATMF would trim values,
regardless of whether or not the 3 × 3 window has an impulse or
2.3 Adaptive median filter none. This possession of UTMF types it extra effective in noise
conquest than the existing ATMF (Jayaraj and Ebenezer 2010).
This filter shall be charity for distinguishing noising and
de-noising pixels after which practice the filtering manner.
A noisy pixel which has been retained by using the median 3 Proposed algorithm: MCF (modified
value and untainted pixels will likely be left unmoved. AMF cascaded filter)
achieves well at low noise concentrations considering the
noising pixels which can be substituted via median values 3.1 Cascaded connection of DMF and UTMF
are only a few. At bigger noise concentrations, window dimen-
sion needs to be accelerated to get improved noise removing DMF is superior to the SMF considering that it gets rid of
that will guide to less association among noising pixel ethics best noising pixels. The drawback of DMF is that, the median
and restored median pixel value (Hwang and Hadded 1995). pixel also be a noisy pixel price outcome in DMF flops to take
away noise with no trouble at excessive density. The UTMF
2.4 Decision‑based median filtering (DMF) has higher noise taking away ability than the existing ATMF
(Luo 2006), as a result of the fact that it uses an un-symmetrical
The primary phase of the modified cascaded filter is the deter- adornment technique to reject only the impulse values. The
mination-headquartered median filter. In normal median filter- UTMF is used to take away the noise thoroughly. At very exces-
ing (SMF), every pixel price is restored through the median sive noise concentrations, around 80% there is obscuring of the
of its regional values (Zhang and Karim 2002). The selection- photograph. The projected cascaded algorithm makes use of a
founded median filter however creates a rough approxima- grouping of both DMF (Lin 1998) and UTMF (Pomalaza-Racz
tion of whether or not a pixel is noising or no longer. Then, and Macgillem 1984), to additional give a boost to the output
the de-noising pixel ethics are retained and the noising pixels obtained from the UTMF. The noisy snapshot is first admin-
were restored by the median value of their neighbours. As a istered using the resolution-headquartered median filter. The
consequence the outcome of DMF is finer to the SMF. output of DMF is given because the center to the UTMF. The
cascaded assembly is consequently used to dispose of salt and
2.5 Un‑symmetric trimmed‑mean filtering (UTMF) pepper noise with a noise concentration as greater than 90%.
Table 1 shows the Noise level and its signal noise ratio (SNR).
The suggestion in the back of clipped mean is to discard The cascaded assembly produces the very best price of PSNR
essentially the maximum likely outliers. Within the current (Chan et al. 2005; Azhar et al. 2019) and IEF (Chen and Wu
alpha-trimmed imply filtering (ATMF), the trimming is sym- 2001; Bisht and Dua 2018) price as shown in Tables 2 and 3.
metric at either end. The quantity of values being clipped is
dependent upon the trimming element “α”. If the value of
α = 4, then “α/2” values are clipped together the ends and the 4 Implementation
remainder pixels are around. It is found that greater the worth
of “α”, bigger is the noise suppression (Ng and Ma 2006). In 4.1 Implementation for color image
ATMF, mostly even un-noising pixel values are additionally
trimmed. This finally results in lack of photo element and The Block Diagram shown in Fig. 1 and algorithm are used
distorting of the image. In to beat the above quandary, an to remove the noise from noisy image using MCF algorithm.
un-symmetric clipped despicable filtering method is planned.
In UTMF (Eng and Ma 2001), the adornment is un-symmet- 4.2 Algorithm (for color image)
ric i.e. The statistics of pixels clipped on the two ends aren’t at
all times equal (Srinivasan and Ebenezer 2007). The UTMF The algorithm for MCF is:
exams whether or not the severe values of the sorted array,
acquired from 3 × 3 window, are impulse values and trim sim- Step 1 2-Dimension window “S × y” of size x = 3, y = 3
plest these impulse values. For illustration, if a 3 × 3 window is chosen.

Table 1  Noise level and its Noise level Very low Low Median High Very high Extremely high
signal noise ratio (SNR)
Signal noise ratio (SNR) ≥ 30 dB ≥ 20 dB ≥ 15 dB ≥ 10 dB ≥ 5 dB < 5 dB

13
Removal of high density salt and pepper noise in color image through modified cascaded filter

Table 2  Comparison of PSNR Noise den- PSNR (in decibels)


values for various filtering sity ND (%)
methods SMF DMF ATMF UTMF α = 2 UTMF α = 4 UTMF α = 6 MCF (PA)

10 33.26 36.35 29.12 31.23 32.41 32.69 32.44


20 28.73 31.41 24.89 28.89 30.68 32.38 32.17
30 23.34 25.35 21.78 25.82 28.43 31.92 31.84
40 18.57 20.43 19.29 22.75 25.98 31.34 31.49
50 15.29 16.72 17.32 19.59 22.83 30.65 31.01
60 12.21 13.31 15.46 17.12 19.79 29.57 30.53
70 10.08 10.89 14.35 14.77 16.61 28.51 29.79
80 8.13 8.67 13.11 12.73 13.42 26.35 28.58
90 6.45 6.87 11.88 11.17 10.76 23.33 26.07

Table 3  Comparison of IEF ND (%) IEF


values for various filtering
methods SMF DMF ATMF UTMF MCF (PA)
α = 2 α = 4 α = 6

10 57.91 128.74 22.74 39.92 50.03 52.12 50.58


20 43.05 79.09 17.96 44.65 66.64 98.26 94.26
30 19.78 29.26 13.05 33.17 59.96 132.88 131.84
40 8.82 12.88 9.88 21.39 44.89 156.19 160.33
50 4.77 6.62 7.56 12.96 27.66 166.02 181.96
60 2.94 3.86 6.18 8.94 16.14 155.77 192.88
70 2.05 2.42 5.34 5.95 9.25 141.69 198.52
80 1.48 1.78 4.52 4.26 5.13 99.25 163.93
90 1.19 1.22 4.08 3.34 3.09 59.16 100.02

Fig. 1  Block diagram for


removal of noise from noisy
color image using MCF algo-
rithm

Step 2 Pixel values in the window are placed in ascending 5 Simulation results
order, and saved in a 1-Dimensional array.
Step 3 If the pixel value in the array is both ‘0’ or ‘255’, The efficiency of the cascaded filter is calculated at specific
the equivalent pixel values are eliminated, and the median phases of noise concentrations in between 10 and 90%. The
of nearer values are considered. outputs are shows in Fig. 1.
Step 4 The pixel being managed is restored by way of the The PSNR value is calculated by
median value analyzed. Transfer the window by using
Step:1, and repeat from Step:2–4. ⎛ ⎞
⎜ 2552 ⎟
PSNR = 10 log10 ⎜ � �∑� �2 ⎟ (1)
Repeat the mention steps, till the process complete for ⎜ ∕M × N
1 rij − xij ⎟
full color image. ⎝ ⎠
Figure 1 Shows the Block diagram for removal of noise
The IEF is calculated, utilizing the system
from noisy color image using MCF algorithm.

13
B. Karthik et al.

COLOR IMAGE FILTERING OUTPUT-MCF ALGORITHM


Original Image Noisy Image SMF DMF ATMF UTMF PA

(a) (b) 30% Noise (c) (d) (e) (f) (g)

(a) (b) 40% Noise (c) (d) (e) (f) (g)

(a) (b) 50% Noise (c) (d) (e) (f) (g)

(a) (b) 60% Noise (c) (d) (e) (f) (g)

(a) (b) 70% Noise (c) (d) (e) (f) (g)

(a) (b) 80% Noise (c) (d) (e) (f) (g)

(a) (b) 90% Noise (c) (d) (e) (f) (g)

(a) (b) 95% Noise (c) (d) (e) (f) (g)

Fig. 2  Lena color image filtering output. a—original image, b—nois- mean filter (ATMF), f—output of un-symmetric trimmed mean filter
ing image, c—output of standard median filter (SMF), d—output of (UTMF), g—output of modified cascaded filter (PA)
decision-based median filter (DMF), e—output of alpha-trimmed

13
Removal of high density salt and pepper noise in color image through modified cascaded filter

Original Image Noisy Image SMF DMF ATMF UTMF PA

(a) (b) 50% Noise (c) (d) (e) (f) (g)

(a) (b) 60% Noise (c) (d) (e) (f) (g)

(a) (b) 70% Noise (c) (d) (e) (f) (g)

(a) (b) 80% Noise (c) (d) (e) (f) (g)

(a) (b) 90% Noise (c) (d) (e) (f) (g)

Fig. 3  Electronic kit color image filtering output. a—original image, trimmed mean filter (ATMF), f—output of un-symmetric trimmed
b—noising image, c—output of standard median filter (SMF), d— mean filter (UTMF), g—output of modified cascaded filter (PA)
output of decision-based median filter (DMF), e—output of alpha-

�∑ �2 However in case of altered cascaded filter there may be an


256
s − rij
i,j=1 ij outright restoration of 90% noise image which will also be
IEF = � �2 (2) observed from the final row under proposed algorithm.
∑256
s − xij
i,j=1 ij Figures 2, 3 and 4 Shows the Color Image Filtering Out-
put-MCF Algorithm.
where s—original image, r—noising image, x—restored In this chapter, a comparative analysis has been made
image, M × N—size of image. among our own proposed schemes on salt and pepper noise
In Figs. 2, 3, 4 the outputs of more than a few filters removal. Both objective as well as subjective evaluations
similarly SMF (Column_3), DMF (Column_4), ATMF are made to derive an overall conclusion. It is observed
(Column_5), UTMF (Column 6), and altered cascaded filter that SMF (Field 1987) and DMF (Mallat 1989) have supe-
(Column_7) are shown. Column 1 symbolizes the common rior performance over others in low density noise (range
photo and column 2 symbolizes snapshot noising by way of Table 1). MCF performs better in both median and high
impulse (salt and pepper) noise (Lee and Bang 2019; Liu density impulse noise. Table 2 shows the Comparison of
et al. 2015). From column 3 SMF may give an equitable PSNR values for various filtering methods.
output best up to 50%, the same factor clenches moral for Figure 5 shows PSNR versus Noise Density is received
DMF and ATMF. An appropriate first-rate of the photograph from the values given in Table 2 for a specified 10% noised
is got for 70% of noised photograph for the duration of the image the maximum PSNR acquired is 36.52 when DMF is
restoration making use of UTMF (Srivastava et al. 2003). used. For 20% noised snapshot the extreme PSNR is 32.35%

13
B. Karthik et al.

(a) – Original Image


(b) – Noising Image
(c) – Output of StandardMedian Filter (SMF)
(d) – Output of Decision-based Median Filter (DMF)
(e) – Output of Alpha-Trimmed Mean Filter (ATMF)
(f) – Output of Un-symmetric Trimmed Mean Filter (UTMF)
(g) – Output of Modified cascaded filter (PA)

(a) (b) (c) (d) (e)

(f) (g)

Fig. 4  Sample color image 50% noise restored using MCF algorithm. output of alpha-trimmed mean filter (ATMF), f—output of un-sym-
a—original image, b—noising image, c—output of standard median metric trimmed mean filter (UTMF), g—output of modified cascaded
filter (SMF), d—output of decision-based median filter (DMF), e— filter (PA)

Fig. 6  IEF versus noise density

Fig. 5  PSNR versus noise density


Figure 6 IEF versus Noise Density had shown a plot of
beneath UTMF (Pok et al. 2003), the pattern gradually alter- image Enrichment factor for quite a lot of filtering algo-
ations in the direction of modified cascaded filter in terms of rithms. Just like the developments of db acquired in Table 3
increased PSNR as the noise densities multiplied as much as here also the IEF is extra for the altered cascaded filter with
90%. Figure 1 i.e., it may be determined the highest PSNR is the photo is compared with SMF, DMF, ATMF and UTMF.
obtained utilizing the modified cascaded filter when in com- Nevertheless the general development is that photos are nois-
parison with other recounted algorithm with noise density ing to the nastiest case and as a result the altered cascaded
extended from 30% ahead. filter will be actual so much apt in eliminating excessive
density noised photograph. Table 3 shows the Comparison
of IEF values for various filtering methods.

13
Removal of high density salt and pepper noise in color image through modified cascaded filter

Fig. 7  Lena image-MCF obtained output for random valued impulse noise

Table 4  Obtained PSNR and IEF


Noise density (%) PSNR (dB) IEF

30 38.9570 25.0468
40 36.5510 17.6718
50 34.2068 13.2275
60 31.4240 9.9905
70 26.4140 7.5376
80 21.3445 5.4451
90 8.4622 3.9973

Fig. 8  Noise density versus PSNR

13
B. Karthik et al.

fractional random transform. J Ambient Intell Humaniz Comput


10(9):3519–3531
Chan RH, Ho C-W, Nikolova M (2005) Salt and pepper noise removal
by median type noise detectors and detail preserving regulariza-
tion. IEEE Trans Image Process 14(10):1479–1485
Chen T, Wu HR (2001) Space variant median filters for the restoration
of impulse noise corrupted images. IEEE Trans Circuits Syst II
Analog Digit Signal Process 48(8):784–789
Eng H-L, Ma K-K (2001) Noise adaptive soft-switching median filter.
IEEE Trans Image Process 10(2):242–251
Field DJ (1987) Relations between the statistics of natural images
and the response properties of cortical cells. J Opt Soc Am A
4(12):2379–2394
Hwang H, Hadded RA (1995) Adaptive median filter: new algorithms
and results. IEEE Trans Image Process 4(4):499–502
Jayaraj V, Ebenezer D (2010) A new switching-based median filtering
Fig. 9  Noise density versuss IEF
scheme and algorithm for removal of high-density salt and pep-
per noise in image. EURASIP J Adv Signal Process. https​://doi.
org/10.1155/2010/69021​8
6 Removal of high density random valued Lee Y-H, Bang S-I (2019) Improved image retrieval and classification
impulse noise through MCF algorithm with combined invariant features and color descriptor. J Ambient
Intell Humaniz Comput 10(6):2255–2264
Lin H-M (1998) Median filters with adaptive length. IEEE Trans Cir-
To remove random valued impulse noise using MCF, noise cuits Syst 35(6):675–690
variance applied from 30 to 90% and resultant output Liu C, Zou H, Li C, Liu Y, Wang Y, Jia S, Zhou S (2015) An adap-
obtained using MCF algorithm. Restoration of 90% noise tive texture-preserved image denoising model. J Ambient Intell
Humaniz Comput 6(5):689–697
image which will also be observed from the final row shown
Luo W (2006) An efficient detail-preserving approach for removing
in Fig. 7 under MCF algorithm obtained output for Random impulse noise in images. IEEE Signal Process Lett 13(7):413–416
Valued Impulse Noise. PSNR and IEF obtained in Table 4. Mallat SG (1989) Multifrequency channel decompositions of images
Figures 8 and 9 PSNR and IEF obtained. and wavelet models. IEEE Trans Acoust Speech Signal Process
37(12):2091–2110. https​://doi.org/10.1109/29.45554​
Ng PE, Ma KK (2006) A switching median filter with boundary dis-
criminative noise detection for extremely corrupted images. IEEE
7 Conclusion Trans Image Process 15(6):1506–1516
Oten R, de Figueiredo RJP (1995) Adaptive alpha-trimmed mean filters
under deviations from assumed noise model. IEEE Trans Image
The restoration of color images which are extremely nois- Process 13(5):627–639
ing by salt and pepper noise and random valued impulse Pok G, Liu J-C, Nair AS (2003) Selective removal of impulse noise
noise. This proposed algorithm replaced the noisy pixel by based on homogeneity level information. IEEE Trans Image Pro-
using clipped median value. This algorithm proved better cess 12(1):85–92
Pomalaza-Racz CA, Macgillem CD (1984) An adaptive non linear
outcome than the Standard Median Filter, Decision Based edge preserving filter. IEEE Trans Acoust Speech Signal Process
Algorithm, Modified Decision Based Algorithm and Pro- ASSP-32:571–576
gressive Switched Median Filter. The modified cascaded Srinivasan KS, Ebenezer D (2007) A new fast and efficient decision
filter (MCF) was checked with various color images and the based algorithm for removal of high density impulse noise. IEEE
Signal Process Lett 14(3):189–192
presentation of the projected filter is loftier to the surviv- Srivastava A, Lee AB, Simoncelli EP, Zhu S-C (2003) On advances
ing filters. The projected algorithm predicted better Peak in statistical modeling of natural images. J Math Imaging Vis
Signal-to-Noise Ratio and Image Enhancement Factor, even 18:17–33
at exact high noise concentrations around 90%, the texture, Zhang S, Karim MA (2002) A new impulse detector for switching
median filters. IEEE Signal Process Lett 9(11):360–363
particulars and edges are conserved efficiently.
Publisher’s Note Springer Nature remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.
References
Azhar M, Dawood H, Dawood H, Choudhary GI, Bashir AK, Chauh-
dary SH (2019) Detail-preserving switching algorithm for the
removal of random-valued impulse noise. J Ambient Intell
Humaniz Comput 10(10):3925–3945
Bisht A, Dua M, Dua S (2018) A novel approach to encrypt multi-
ple images using multiple chaotic maps and chaotic discrete

13

You might also like