A SURVEY OF FILTERING TECHNIQUES FOR IMAGE …ijrpublisher.com/gallery/27-february-2019.pdf · 2019. 2. 8. · In recent years most of the filtering based techniques are based on

A SURVEY OF FILTERING TECHNIQUES FOR

IMAGE ENHANCEMENT

1Dr.N.Nandhagopal,

2V.Nivedita,

3R.Kanagaraj,

4S.Theivanayaki

1Associate Professor,

2Assistant Professor,

3Assistant Professor,

4Assistant Professor

1,3,4Electronics and Communication Engineering, Excel Engineering College, Namakkal, India

2Computer Science and Engineering, Star Lion Engineering College, Thanjavur, India.

Abstract: Image filtering is the process that enables us to do some modifications or apply filters on pictures.

Typically the filtering or filter is a thing via which a thing is passed to oust the unwanted particles contained

in it. Here the image filtering is done when we obtain the picture from camera or other device we suppose to

do some effects or modifications on it through image filtering techniques for mitigating the blur or

unwanted substances like noise on the image to enhance the visual quality of the image. The image filtering

is also known as image smoothing which is one of the most eminent process and widely employed for the

image processing ones. The scope of smoothing is to reduce blur and enhance the visual clarity of the

particular image by employing diverse filtering techniques. The most typically used algorithms for filtering

images are linear algorithm and non linear filtering algorithms. The non linear filter is quite distinct

characteristics while compared to linear filter because the response of the given input will not adhere the

principles specified earlier. Filtering becomes an substantial method of signal processing system. It includes

the degradation of signal and its performance. In recent years most of the filtering based techniques are

based on linear processing methods. In this paper we survey and compare the various image filtering

techniques for image processing system. The image filtering techniques are Gaussian filter, Median Filter,

Average Filter, Spatial Filter, Fuzzy Filter, Adaptive Filter, etc. These filters are categorized as Linear and

Non Linear Filters.

Index Terms – Image Filtering, Smoothing, Linear Filter, Non Linear Filter.

I.INTRODUCTION

In recent years there is a vast amount of compact or portable devices coming in image processing and

multimedia applications which are smart phones, camera, tablets, wearable’s, gadgets, etc. These smart

devices typically have restrained computing, power consumption, and storage services which are demanding

high latency sensitive applications. So that the techniques of energy efficient methods are becoming the vital

one for mitigating the power consumption of resource constrained devices. There are several methods to

attain energy efficient methods in the architecture. The most of the image processing or filtering

International Journal of Research

Volume VIII, Issue II, February/2019

ISSN NO:2236-6124

Page No:180

applications gives output for the human assumption that shows restrained view and the minor deviation in

the output image could not be identified. This application which we typically called as fault tolerant

application that is minor deviations in the output is reasonable. Filtering is often called smoothing which

removes the unwanted noise and blur in the particular image and surge the high resolution of the image by

applying appropriate filters such as Guassian Filter, Fuzzy Filter, FIR Filter, Median Filter, and Adaptive

Filter, etc.

Image Filtering

The term convolution has been broadly used in computer vision and image filtering or processing

techniques which comprises things or entity recognition and picture mapping. Though convolution process

commonly entails a substantial amount of high intensive computing resources such as CPU, Memory, etc.

The image filtering process is employed as preprocessing that is to remove redundant things, noise, and blur

from image. Filtering is the process that the data that is originated from smart devices such as tablets,

phones, etc are preprocessed or filtered through computing technologies. In the literature various FPGA

prototypes are implemented in the 3D Convolution method. The 2D Gaussian function of the image filtering

application renders a savings of computing resources. The filtering output image is recognized by

employing some algorithm to the data in the pixels. The neighborhoods play a essential role in advanced

digital image processing systems. It is therefore substantial to realize how snaps can be sampled and how it

pertains to several neighborhoods that could be utilized to process the particular image. The types of

neighborhoods are

a) Rectangular Sampling.

b) Hexagonal Sampling..

Rectangular Sampling

In Rectangular sampling the images are preserved or sampled by having in a rectangular part over an image.

The Figure 1. It Is the type of Rectangular Sampling.

Fig. 1 Rectangular Sampling



ISSN NO:2236-6124

Page No:181

Hexagonal Sampling

The Hexagonal sampling is the alternative or optional one to the Rectangular Sampling. The Figure 2. It Is

the type of Hexagonal Sampling. Here the signal is circularly bandwidth limited in the same

communication frequency in the same resolution square image. The Hexagonal Sampling reduces the

various points of sampling by a factorization method of 2/3(1/2).

Fig. 2 Hexagonal Sampling

Characteristics of image operations

In image operations there are several ways to categorize and characterize operations of image. The rationale

is to construe what type of response we get from the given input value to attain a specific operation or what

could be the complex problems associated in a given operation. The types of image operations can be

explored in digital images to translate an input image to an output image. It is categorized in to three

operations based on their characterization.

a) Point – the input and output value in a particular coordinate is dependent on the only input value in

the same coordinate. And its generic complexity or pixel is constant.

b) Local - the input and output value in a particular coordinate is dependent on the only input value in

the neighborhood same coordinate. And its generic complexity or pixel is p^2.

c) Global - the input and output value in a particular coordinate is dependent on the all input values in

the input and output coordinate. And its generic complexity or pixel is N^2.



ISSN NO:2236-6124

Page No:182

Fig. 3 Point, Global, Local Image Operations

Image Filtering Algorithm

The image filtering algorithm also called software image filtering that process the image or snap to bypass a

specific set of frequency components. When maps to audio, reflection, and communication the word

frequency is measured is cycle per second. To understand the patterns as component frequency in images

the image filtering is often used to filter the images to obtain the high definition content without noise, blur,

and unwanted things. The filtering is also used to improve the patterns of spatial determined by the intensity

of light instead of light frequency.

Fig. 4. Block diagram of Image Filtering



ISSN NO:2236-6124

Page No:183

The image consists of collection of component frequencies. To determine how filtering process happens it

can prototype the imaging functions such as pass, adaptive, median, and various other frequencies. The

scope of image filtering is to identify the needed data in certain levels of frequency spectrum to expel the

unwanted frequencies. In image filtering algorithm smoothing filters are applicable in many applications are

object sensing, mapping, and categorization, etc. These are applied for the preprocessing the redundant

details and noise. There are two algorithms are,

a) Gaussian Mask.

b) Convolution Operation

Gaussian Mask

Guassian Mask also called guassian filter which is one of the prominent and broadly deployed filtering

algorithms in image processing or filtering techniques. The Guassian Filter (G) is defined as G(x,y) = ½

(pi)(e^(x+y)). Where G is the Gaussian Filter, x and y are the coordinates. The Guassian Mask that uses a

Guassian Function for expressing the normal distribution in the formula and also for reckoning the

transformation to employ to every pixel in the image identified if not the procedure recurs again to extract

the matched eyes from the millions of records stored in the database. The rationale is to construe what type

of response we get from the given input value to attain a specific operation or what could be the complex

problems associated in a given operation. The types of image operations can be explored in digital images to

translate an input image to an output image. It is categorized in to three operations based on their

characterization. The image filtering process is employed as preprocessing that is to remove redundant

things, noise, and blur from image. Filtering is the process that the data that is originated from smart devices

such as tablets, phones, etc.

Fig. 5. Guassian Mask



ISSN NO:2236-6124

Page No:184

Convolution operation

The Convolution operation is a imaging process that process a multiple shift integration operation. During

the run time the pixel of each image is integrated or centralized over other pixels in an old image. The

earlier values of pixel under the Operating system kernel is multiplied by the associated processing pixels

for example all are same as 1. The outcomes of all multiplications are integrated in to one image and the

outcome values are reported in the output image. The kernel processing is for calculating the simple

mathematical methods for many typical image processing operations. In the image filtering context it

renders a way to separate the two arrays of numbers in the equal dimensions which can be utilized in image

filtering techniques. The output pixel values are basic nonlinear and linear associations of some input pixel

values.

Generally smoothing can be designed by complying the original image of taken picture of the size x and y

includes Gaussian mask. It is acquired by calculating the sum of products of the input and output image with

the matrix of the size 3*3 using the convolution operation. In convolution linear filtering can be applied

using a method called discrete convolution. Ceaseless convolution is typical in image and signal processing

systems but the snaps or images are not discrete. Here we employ only discrete convolution.

a) Specifies the weight of pixel as an image, represents as K.

b) K the image representation commonly called kernel in convolution.

c) Here the kernel operation is associative.

The earlier values of pixel under the Operating system kernel is multiplied by the associated processing

pixels for example all are same as 1. The outcomes of all multiplications are integrated in to one image and

the outcome values are reported in the output image. The kernel processing is for calculating the simple

mathematical methods for many typical image processing operations. The scope of image filtering is to

identify the needed data in certain levels of frequency spectrum to expel the unwanted frequencies. In image

filtering algorithm smoothing filters are applicable in many applications are object sensing, mapping, and

categorization, etc.

Fig. 6. Convolution Operation



ISSN NO:2236-6124

Page No:185

II. IMAGE FILTERING TECHNIQUES

Guassian Filter

The Guassian Filter is one of the widely used image filtering techniques and also it is denoted as weighted

non linear filter which is employing for ousting the blur and noise of the image. The Guassian Filter is

typically utilizing for the preprocessing method in the particular image before we go for the edge detection

to oust the redundant particles or entities caused by noise. In order to efficiently utilize or implement GF

(Gaussian Filter) the equation of Guassian is determined by the operating system kernel of variant sizes. The

larger size of OS Kernel is good in the Gaussian expression or equation which renders high definition

quality for viewing the image visually in the lower cost computational complexity. To attain the filtered

image the convolution operation is performed with the Guassian Kernel along with sub matrix image which

exhibits constant multiplier floating point since it consumes high energy.

Energy Efficient Filter Design

In Energy efficient Filter design we can review the variance of energy efficient operating system kernels for

efficiently calculate the filtered pixel. The following energy efficient kernels are,

a) Fixed Point Guassian Kernel.

b) Digital Approximated 2D Gaussian Kernel.

c) Low complexity using neighbor pixels similarity.

d) Approximate Gaussian Filter using approximate adders.

Fixed Point Guassian Kernel

There Fixed point Guassian Kernel is one of the energy efficient operating systems kernel. It has developed

mainly for mitigating the implementation complexity of coefficients of kernel to the static point rather than

dynamic ones. In static point configuration the format (1,m) specifies the number of bits or bytes. The m

denotes the location of the least significant bits. The operating systems kernel mitigate the complexity of

employing the filter which produces the outcomes in substantial reduction in energy consumption and delay

metrics. And also the kernel provides the unacceptable quality delay.



ISSN NO:2236-6124

Page No:186

Fig. 7. Fixed Point Guassian Kernel

Digital approximated 2D Guassian Kernel

The digital approximated 2D Kernel which substantially mitigates the Guassian Kernel implementation

complexity. In Guassian Kernel each coefficient is determined by sum and power of two. Since the

multiplication of any value or term with a static value in power of two does not require hardware and

software implementation. The architecture that utilized the coefficients of kernel is the sum and power of

two to calculate the smoothened pixel. The quality metrics are as follows,

a) Mean Square Error (MSE).

b) Normalized Error Distance (NED).

c) Peak Signal to Noise Ratio (PSNR).

d) Structural Similarity (SSIM).

In order to efficiently utilize or implement GF (Gaussian Filter) the equation of Guassian is determined by

the operating system kernel of variant sizes. The larger size of OS Kernel is good in the Gaussian expression

or equation which renders high definition quality for viewing the image visually in the lower cost

computational complexity. The kernel processing is for calculating the simple mathematical methods for

many typical image processing operations. In the image filtering context it renders a way to separate the two

arrays of numbers in the equal dimensions which can be utilized in image filtering techniques.



ISSN NO:2236-6124

Page No:187

Fig. 8. Digital Approximated 2D Guassian Kernel

Low Complexity using Neighbor Pixels Similarity

In low complexity using neighbor pixels similarity the neighbor pixel in each image shows correlation is

higher that is dta or information in image pixels will be the same nearly. This method or property could be

explored to mitigate the implementation complexity of the low complexity using neighbor pixels similarity.

Since most of the images have the higher visual quality the edge of images is very tiny in number which

assumes that adjacent pixel of every images is nearly same. In case of smoothing the image the sub matrix is

processed for to acquire the better quality of image. So that in context of neighbor pixel should be the same

as like neighbor pixel similarity (NPS) when applied to sub matrix which significantly mitigate the burden

of computational complexity of smoothening the image filter. The low complexity neighbor pixels similarity

comprises of three common steps which are as follows,

a) Image preprocessing

b) Feature detection

c) Ellipse fitting

Fig. 9. Low Complexity using Neighbor Pixels Similarity



ISSN NO:2236-6124

Page No:188

Approximate Guassian Filter using Guassian Adders

In Approximate Guassian Filter using Guasssian Adders where approximate adders are physically

embedded for the precise GF (Guassian Filter) to barely mitigate the noise, area, latency, and energy. In the

approximate Guassian Filter using Approximate Adders various bit – width adders are exploited to attain

energy and power consumption trade off. The coefficients of kernel is to prototype or design the

approximate filter. The adders which are approximate one of the approximate adders are error tolerant

adders which are exploited to attain power consumption and energy efficiency. The architecture of

approximate Guassian Filter GF is determined for viewing the better visual quality image. Here the k value

denotes the how many bits in adder in which approximate logic is employed to reckon the sum, further this

architecture entails low number of adders for approximating or reducing the implementation complexity.

The various bit – width calculations are implemented to acquire the enhanced metrics of design and quality.

This method or property could be explored to mitigate the implementation complexity of the low

complexity using neighbor pixels similarity. Since most of the images have the higher visual quality the

edge of images is very tiny in number which assumes that adjacent pixel of every images is nearly same. In

case of smoothing the image the sub matrix is processed for to acquire the better quality of image. So that in

context of neighbor pixel should be the same as like neighbor pixel similarity (NPS) when applied to sub

matrix which significantly mitigate the burden of computational complexity of smoothening the image filter.

the techniques of energy efficient methods are becoming the vital one for mitigating the power consumption

of resource constrained devices. There are several methods to attain energy efficient methods in the

architecture. The most of the image processing or filtering applications gives output for the human

assumption that shows restrained view and the minor deviation in the output image could not be identified.

Fig. 10. Guassian Adders

Median Filter

The Median Filter is one of the second image filtering techniques. The Median Filter is comes under the non

linear digital image filtering technique or method. The median filter technique is mostly used for expelling

noise or blur from the signal or image. This reduction of blur or noise is a common filtering or

preprocessing steps to enhance the outcomes of subsequent processing. We can denote the example is



ISSN NO:2236-6124

Page No:189

detection of edge in an image. The median filtering is broadly exploited in digital image processing

technique since under some laws or conditions the median filtering preserves edges on the particular image

when mitigating the noise and also possess applications in digital or signal processing. The boundary issues

in median filtering are,

a) Dodge the boundaries processing with or without adding the signal and image boundary conditions.

b) Fetching the attributes related to images in the signal. The images in example entries from the

boundaries of horizontal or vertical elements could be selected.

c) Shrinking the boundaries near the window such that evry window is complete.

Fig. 11. Median Filter

Objective of Median Filtering

The objective of median filtering technique is mostly used for expelling noise or blur from the signal or

image. This reduction of blur or noise is a common filtering or preprocessing steps to enhance the outcomes

of subsequent processing. The objectives are as follows,

a) To exclude signal weakening which includes counters of objects and edges blurred.

b) Make sure that better image pixels are in intact irrespective of volume and density of noise in the

picture.

c) Abstain the circumstance where the recognized noise pixel is replaced with some other noise

pixel.

d) Mitigate the algorithm’s time complexity.

e) Facilitate the median filtering to sense and preserve details of redundant noise in images.



ISSN NO:2236-6124

Page No:190

Drawbacks

The drawbacks of the Median Filtering are,

a) Senses noise and replace with another noise pixel.

b) Affecting some of the good pixels when volume or density of the nose is low.

c) Surging the processing time while the density of the noise in the picture is high.

d) Hard to sense and preserve edge details in an image.

Fuzzy Filter

One of the most popular methods of image filtering is Fuzzy filter. It renders outcomes in image or digital

processing tasks that deal with some shortcomings of the classical filters. Fuzzy filter is potential of arguing

with the vague and unsure data or information. Most of the times it entailed to rejuvenate a heavy noise

polluted image in which there is a lot of unwanted particles present in image. Every pixel in the image is

denoted by a function membership and various types of rues in fuzzy filters that points out the information

of neighborhood to remove the filter with blurred images.

Fig. 12. Fuzzy Filter

Adaptive Filter

The Adaptive Filter is also called Wiener Filter and it comes under the category of linear filters. Here if the

difference is large weiner performs bit smoothing technique. This is one of the approach adaptive filter often

renders outcome better than the linear filtering. The adaptive filter is more precise than a non linear and

linear filter in the way of preserving edges also better frequency parts in an image. Therefore there are no

prototype tasks the weiner function directs all the fundamental computations and implements the adaptive



ISSN NO:2236-6124

Page No:191

filter for the particular input image. Adaptive filter requires more processing or computation time than the

non linear filtering. The weiner works better in case of expelling noise in image.

Fig. 13. Adaptive Filter

Spatial Filter

The Spatial Filter is one of the image filtering techniques. It is the optical device which employs the laws of

fourier optics to transform the structure of light beam and other electromagnetic radiation. It commonly

called coherent laser light. The spatial filtering is typically used to erase or clean up the blur or noise of the

images also expel abbreviations in the light beam because of imperfect, noise, blur, and other damaged

optics. The spatial filtering can be applied to send a transverse mode from multiple lasers, when blocking

other modes reflected from the light. The term spatial filtering indicates the desirable common features od

the original image. The applications of spatial filtering which are as follows,

a) Sensing and sharpening boundary unceaseless images for the fingerprints, remote sensing images.

b) Expel unwanted noise from a laser beam

c) Locate the faults or errors in the masks used to create integrated circuits.

d) Removing unwanted characteristics from the photographs or images.

e) Usability research

Classification of Filters

The term filtering has been procured from the frequency domain. In spatial filtering filters are classified as

follows,



ISSN NO:2236-6124

Page No:192

a) Low Pass ( Preserve low frequencies )

b) High Pass ( Preserve high frequencies )

c) Band Pass ( Preserve frequencies within a band )

d) Band Reject ( Reject frequencies within a band)

Fig. 14. Spatial Filter

III.CONCLUSION

This paper has presented an survey of image filtering techniques by comparing various filtering techniques

such as spatial filter, adaptive filter, fuzzy filter, median filter, and Gaussian filter. There are several

methods to attain energy efficient methods in the architecture. The most of the image processing or filtering

applications gives output for the human assumption that shows restrained view and the minor deviation in

the output image could not be identified. This application which we typically called as fault tolerant

application that is minor deviations in the output is reasonable.

IV. REFERENCES

[1] Harsh Prateek Singh, A. 2014. Noise Reduction in Images using Enhanced Average Filter. International

Journal of Computer Applications, pp. 16 – 19.

[2] Lan Huang, Chiou-Shann Fuh; Noise Reduction Using Enhanced Bilateral Filter, Department of

Computer Science and Information Engineering, National Taiwan University, Taipei, Taiwan, 10617,

R.O.C,2006,vol 12,No 4



ISSN NO:2236-6124

Page No:193

[3] Shyam Lal1, Sanjeev Kumar 2and Mahesh Chandra3:Removal of High Density Salt & Pepper Noise

Moradabad-244001(UP), India 2,3, ECE Department, Birla Institute of Technology,Mesra,Ranchi-

835215(JH),India.

[4] F. Russo and G. Ramponi, “A fuzzy operator for the enhancement of blurred and noisy images,” IEEE

Trans. Image Processing, vol. 4, pp. 1169–1174, Aug. 1995.

[5] R. C. Gonzalez and R. E. Woods, Digital Image Processing, 2nd Ed., Prentice-Hall, Upper Saddle River,

NJ,2002.

[6] Amer Al-Rahayfeh And Miad Faezipour, D, 2013. Eye Tracking and Head Movement Detection:A

State-of-Art Survey. IEEE Transitional Enginerring and Health Medicine, pp. 1536-1284.

[7]C. Tomasi, R. Manduchi, "Bilateral Filtering for Gray and Color Images," Proceedings of International

Conference on Computer Vision, Bombay, pp. 839-846, 1998.

[8] K.Rajiv Gandhi, N.Nandhagopal, R.Sivasubramanian, “Automatic System For Pre-Processing And

Enhancement Of Magnetic Resonance Image (MRI)”, International Journal of Applied Engineering

Research (IJAER) vol.9 (22),pp. 15485-15499, 2014.

[9] Jaspreet Kaur,Salt & Pepper Noise Removal using Fuzzy Based Adaptive Filter; International Journal of

Science, Engineering and Technology Research (IJSETR) Volume 1, Issue 1, July 2012 .

[10] Ana Isabel Molina , Óscar Navarro B , C , Manuel Ortega A , Miguel Lacruz C, 2018. Evaluating

multimedia learning materials in primary education using eye tracking . Computer Standards and Interfaces,

pp. 978 – 916.

[11] V.Velusamy, Dr.M.Karnan, Dr.R.Sivakumar, Dr.N.Nandhagopal, “Enhancement Techniques and

Methods for MRI A Review”, International Journal of Computer Science and Information Technologies,

Vol. 5 (1), pp .397-403, 2014.

[12] H.D. Cheng and X.J. Shi,”A simple and effective histogram equalization approach to image

enhancement”, Department of Computer Science, Utah State University, Logan, UT 84322-4205, USA.



ISSN NO:2236-6124

Page No:194

A SURVEY OF FILTERING TECHNIQUES FOR IMAGE …ijrpublisher.com/gallery/27-february-2019.pdf · 2019. 2. 8. · In recent years most of the filtering based techniques are based on

Documents