Review of Various Techniques for Removing Impulse Noise in Digital Images

Review of Various Techniques for Removing Impulse Noise in Digital Images

AnshumanTyagi, PriyaVarshney

CSE Department, Pranveer Singh Institute of Technology, Kanpur

CSE Department, Pranveer Singh Institute of Technology, Kanpur

Abstract

The quality of an image is reduced due to the presence of noise.Noise hides the important details of images.There are various types ofnoise;Impulse noise is one of them. The principal source of noises in digital images arise during image acquisition and/or transmission,due to light levels,due to sensor temperature, switching due to faulty scanner,due to the interference in the channel or other atmospheric disturbance. For enhancing the images varioustechniques are used to remove noise while preserving the various details of the images.In this paper we have presented the concise overview of various techniques used for removing the impulse noise in digital images and their limitations.

Keywords: Impulse Noise,Detail-preserving, image restoration, impulse noise detection, fuzzy

1. Introduction

Noise removal is one of the major part in the field of image processing .Removing impulse noise can be an ultimate goal to improve the visual quality of images .The goal of removing impulsive noise is primarily to suppress the noise as well as to

preserve the integrity of edges and detailed information like structural features, textural information. Impulse noise corruption is very common in digitalimages[11]. Impulse noise is always independent and uncorrelated to the image pixels and is randomly distributed over the image. Hence

unlike Gaussian noise,for an impulse noise corrupted image all the image pixelsare not noisy, a number of image pixels will be noisy andthe rest of pixels will be noise free. There are differenttypes of impulse noise namely salt and pepper type ofnoise and random valued impulse noise.

In salt and peppertype of noise the noisy pixels takeseither salt value (gray level -225) or pepper value (greylevel -0) and it appears as white and black spots on theimages.If p is the total noise density then salt noise andpepper noise will have a noise density of p/2.This can bemathematically represented by (1)

Zero or 255 with probability p

yij=

xijwith probability 1-p (1)

where yijrepresents the noisy image pixel, p is the total noise density of impulse noise and xij is the uncorruptedimage pixel. At times the salt noise and peppernoise mayhave different noise densitiesp1andp2and the total noisedensity will be p=p1+ p2 .

In case of random valued impulse noise, noise can take anygray level value from zero to 225. In this case also noise israndomly distributed over the entire image and probabilityof occurrence of any gray level value as noise will be same.

We can mathematically represent random valued impulsenoise as in (2).

nijwith probability p

yij=

xijwith probability 1-p (2)

wherenijis the gray level value of the noisy pixel.

1. Literature Review

There are various techniques to remove the impulse noise from images. In this paper we study various techniques provided by researchers to enhance the quality of image by removing the impulse noise.

2.1Mean Filter

Mean filtering [2] is a linear filter and easy to implement method of smoothing images, i.e. reducing the amount of intensity variationbetween one pixel and the next. It is often used to reduce noise in images. The idea of mean filtering is simply to replace each pixel value in an image withthe mean (`average') value of its neighbours.This has

the effect of eliminating pixel values which areunrepresentative of their surroundings.

Drawback and limitation of Mean filter are as follows:

1.Single pixel with a very unrepresentative value can significantly affect the mean value of all the pixels in its neighborhood.

2.When the filter neighborhood straddles an edge, the filter will interpolate new values for pixels on the edge and so will blur that edge. This may be a problem if sharp edges are required in the output.

2.2 Weiner Filter

The wiener (Minimum Mean Square Error) [2] filtering is a linear filter method requires theinformation about the spectra of the noise and theoriginal signal and it works well only if the underlying signal is smooth. Wiener methodimplements spatial smoothing and its modelcomplexity control correspond to choosing thewindow size. Wiener filtering is able to achievesignificant noise removal.

Drawback and limitation of Weiner filter are as follows:

1. When the variance of noise is low, they cause blurring and smoothening of the sharp edges of the image.
2. Spatially invariant

2.3 Median filter

The median filter [3]is a nonlinear digital filter which is often used in digitalimage processing to reduce noise in an image. Inpractice, besides reducing noise, it is important topreserve the edges of an image as edges providecritical information on the visual appearance of animage. Median filtering is a smoothing techniquewhich is effective in reducing noise in the smoothregions of an image. But can adversely affect thesharpness in edges. For small to moderate levels ofsalt and pepper noise the median filter has shown tobe useful in reducing noise while preserving edges.

A major advantage of the median filter over linear filters is that the median filter can eliminate the effect of input noise values with extremely large magnitudes.

Drawback and limitation of Median filter are as follows:

1. With deteriorating performances at a highlevel of noise.
2. Not suitable for high noise densities anddoes not preserve the image details likeedges for further post processing.

2.4 Weighted Median Filter

WM filter have the robustness and edge preserving capability of the classical median filter. WM filter [4] is much more flexible in preserving desired signal structures than a median filter. Edge preservation is essential in image processing due to the nature of visual perception. The most commonly used one assumes positive integer weights with odd sum. WM filters were investigated under several typical structural constraints: line preservation, area preservation and compound details preservation. Median filters, however, often blur images when window becomes larger. Weighted median filters are, then, proposed to solve this problem [3]. Weighted median filters, when properly designed, can preserve finer image details than the standard median filter under the same noise attenuation. Weights may be adjusted to yield “best” filter. An N-length WM filter can be described by N parameters and implemented using a sorting operation with the same order of computations as the same size median filter. On the other hand, WM filters offer much greater flexibility in design specifications than the median filter.

2.5 Adaptive Median Filtering

The Adaptive Median Filter [5] performs spatial processing to determine which pixels in an image have been affected by impulse noise. The Adaptive Median Filter classifies pixels as noise by comparing each pixel in the image to its surrounding neighbor pixels. The size of the neighborhood is adjustable, as well as the threshold for the comparison. A pixel that is different from a majority of its neighbors, as well as being not structurally aligned with those pixels to which it is similar, is labeled as impulse noise. These noise pixels are thenreplaced by the median pixel value of the pixels inthe neighborhood that have passed the noise labeling test.The advantage of adaptive median filter over standard median filter is that the standard median filter does not perform well when impulse noise is Greater than 0.2, while the adaptive median filter can better handle these noises.This filter is to be robust in removing mixed impulses with high probability of occurrence while preserving sharpness. The application of adaptive median filter is communication, radar, sonar, signal processing, interference cancellation, active noise control, biomedical engineering [9].

Drawback and limitation of Adaptive Median Filtering filter are as follows:

1. Above approaches might blur the imagesince both noisy and noise free pixels aremodified.
2. Existing systems uses fixed or differentwindow size for detection of impulse noise.No algorithm is exist which can remove thenoise from the edges of the gray scaleimage.

2.6 Fuzzy Filter

Fuzzy filter uses the concept of fuzzy logic for filtering the images.Fuzzy logic represents a good mathematical framework to deal with uncertainty of information.Fuzzy filters eliminate impulse noise satisfactorily.This uses two separated steps: an impulse noise detection step and reduction step. Fuzzy based filters [6] can also be furtherclassified into two categories: fuzzy classical and fully fuzzy. Fuzzy classical filters include the filterswhich extend the traditional filters using fuzzy logic.There are plenty of fuzzy traditional filters on whichmany researchers have worked. We here mentiononly some of them. Popular fuzzy classical filters are:Fuzzy Median Filter (FMF), Fuzzy Impulse noiseDetection and Reduction Method (FIDRM), FuzzyRandom Impulse Noise Reduction method (FRINR),Fuzzy Weighted Mean (FWM), Adaptive WeightedFuzzy Mean (AWFM).This filter reduces distortion like excessive thinning or thickening of object boundaries.

2.7A Detail Preserving Fuzzy Filter

There is a tendency for any impulse detection scheme to misclassify the edge pixels in the corrupted image as noise and vice versa, since the nature of the noise and edge in an image appear to be similar due to their sudden transition in the gray level value [7]. It becomes imperative to differentiate between noisy and edge pixels. Extracting the edges from the corrupted image is also a difficult task without having knowledge about the edge information. The issue of median filter removes both the noise and the fine details such as thin lines, sharp corners, textures since it can’t tell the difference between the two. If the image is affected only by low level noise, weighted median filter preserve edges.Identifying noise free, noisy and edge pixels still looms large, which has a direct implication on reserving details in an image during filtering.

2.8Noise detection and Removal Using

Cellular Automata:

In this algorithm [8], the first step is the detection ofnoisy points will be done using Adaptive BoundaryDifferentiate Noise Detection Method by CellularAutomata. In the second step, the value ofdetected noisy pixels will be changed accordingto a Switching Median Filter based on CellularAutomata, which is used to remove impulse noisefrom noise corrupted images. In this algorithm an efficient Switching Median Filter based onCellular Automata are proposed to remove low tohigh value of salt and pepper noise. Theproposed filter performs well for gray scale imagewith different noise model of salt and pepper noise.In this algorithm Adaptive boundary DiscriminativeNoise Detection can accurately tell where the noiseis, only noise affected pixels are replaced bySwitching Median Filter based on CellularAutomata and intensity of unaffected pixelswithin the working window. Cellular Automata isa simple and robust method with parallelizationability used in for image processing applicationssuch as improving image quality and it is best suitedfor real time applications.

Drawback and limitation of Noise detection and Removal Using Cellular Automata Median Filtering are as follows:

However, despite its simplicity, thismethod like median filter or other noisereduction methods, delivers a good resultonly when the corrupted noises are low, butin case of a high noise occurrence alleffective pixels will change.

2.9 A Novel Recursive Algorithm:

Given an arbitrary impulse noise removal filter withseparable noise detection and correction operations,denoted by NoiseMapand Estimate, respectively [9]. The noise map of a noisy image is a binary image with the samesize as the noisy image where foreground and background pixelsrepresent noisy and noise-free pixels, respectively. The aim is to enhance the filter’s performance byapplying the proposed recursive algorithm, whichincludes two main steps. In the first, an initial set ofnoisy pixels is identified, which is called the noise

map. Then, these noisy pixels are recursivelycorrected based on the number of noise-freeneighbors. Since the initial noise detection is nonrecursive,the noisy neighbors of each correctednoisy pixel are re-examined to verify whether theyare still noisy or not.The main aim of this technique is to maximize the contribution of noise-free neighbours in detecting and correcting noisy pixels.Unlike the classical recursive implementation which performs sequential row-by-row scanning, this algorithm maximizes the contribution of noise-free neighbors in detecting and correcting the noisy pixels. Regions with lower noisy density are filtered before regions with higher noise density. This algorithm recursively corrects the noisy pixels of an image.Unlike the classical median filter [5], the detection stage avoids altering noise-free pixels which results in preserving the image details.

2.10 A Fusion Technique:

Imagefusion is the process of combining two or more images intoa single image while retaining the important features ofeach image. Multiple image fusion[10] is an importanttechnique used in military, remote sensing and medicalapplications. Five different filtering algorithms are usedindividually for filtering the image captured from thesensor. The filtered images are fused to obtain a highquality image compared to individually denoised images. The fusion technique is shown in figure1.

The image captured by thesensor undergoes filtering by different smoothing filtersand the resultant images are fused to attain high qualityimage. This restoration of image data is very likely to findpotential applications in a number of different areas such aselectromagnetic imaging of objects, medical diagnostics,remote sensing, robotics, etc.

2.11 A Filtering Technique based on average absolute value of four Convolutions:

Anew technique forimpulsedetectionandfilteringbasedontheaverageabsolutevalueoffourconvolutionsobtainedbyone-dimensionalLaplacianoperators is used to remove the impulse noise[12]. Therearethreesteps inalgorithmforimpulsedetectionandfiltering.Afterclassifyingcorruptedanduncorruptedpixel,wereplacethecorruptedpixelbythesuitablevalueofthesortedsequenceofitsneighborhoodvalue.Werepeat these threestepstogettheconvergentrecoveryimage.

Step(a):TheinputimageXijisfirstconvolvedwithasetofconvolutionkernels. Then,theaverageabsolutevalueofthesefourconvolutionsisusedforimpulsedetection

Step(b): Identify whether the pixel Xij is noisy or noise free pixel

Step(c): If pixel Xij will be regarded as noisy pixel,Xij isreplacedbythe meanofnoisefreepixels.

Itisobservedthatthisalgorithmcanprovidebetterperformanceintermsofimagequalityandcomputationtime. This algorithmcaneffectivelyremovethe

impulsenoisewithawiderangeofnoisedensityandproducebetterresultsintermsofthequalitativeandquantitativemeasuresoftheimagesevenatnoisedensityashighas90%.

1. SUMMARY TABLE

1. CONCLUSION

In this paper different impulse noise removal techniques are studied and compared. This paper gives the review of various methods used for filtering the impulse noise of digital images and merits and demerits of these methods. The mean filter is easy and simple linear technique to remove impulse noise but problem may arise if sharp edges are required in the output. The weinerfilter achieve noise removal when the variance of noise is low but they cause blurring and smoothening of the sharp edges of the image. The standard median filter is non linear technique and has advantage over linear filters is that it caneliminate the effect of input noise values with extremely large magnitudes, but not suitable for high level of noise. Weighted median filters, when properly designed, can preserve finer image details than the standard median filter under the same noise attenuation. The adaptive median filter can better handle impulse noise greater than 0.2.The Fuzzy filter reduces distortion like excessive thinning or thickening of object boundaries.A Detail preserved fuzzy filters identify noise free, noisy and edge pixels which has a direct implication on reserving details in an image during filtering.Cellular Automata isa simple and robust method with parallelizationability used in for image processing applicationssuch as improving image quality and it is best suitedfor real time applications.Unlike the classical median filter, Novel Recursive filter detects the noisy pixel, the detection stage avoids altering noise-free pixels which results in preserving the image details. In the lastalgorithm, itcaneffectivelyremovetheimpulsenoisewithawiderangeofnoisedensityandproducebetterresultsintermsofthequalitativeandquantitativemeasuresoftheimagesevenatnoisedensityashighas90%.