COLORED IMAGE RESTORATION USINGNEIGHBORING PIXEL METHODNazia HossainDepartment of CSEMilitary Institute of Scienceand Technology, Dhaka, [email protected] HaqueDepartment of CSEMilitary Institute of Scienceand Technology, Dhaka, [email protected]
comNaresh Sing ChauhanDepartment of CSEMilitary Institute of Scienceand Technology, Dhaka, [email protected]—The field of image processing is evolving rapidly.During the recent years, there has been a significant improvementin the level of interest in image processing algorithm forimage restoration, image registration, image morphology, neuralnetworks, full-color image processing, image data compression,image recognition and knowledge-based image analysis systems.An image can be corrupted by various kind of motion, noise ordistorted signal during the process of acquisition and to detectthe reason for distorting an image various method of restorationhas been introduced in image processing. To restore a corruptedimage there are number of methods have been introduced inthis eld including neighboring pixel method. Under this context,an effort has been given to restore a corrupted image with theneighborhood pixel method using mean value. We calculated themean values of neighbor pixel and replaces the mean valueswith the target pixel of input image.
We investigated the noiselevel of RGB and grayscale images and improved the noise levelusing neighboring pixel method that enhances the target imagecomparing with existing noise reduction filters.Index Terms—Image processing, Neighboring Pixel, ImageRestoration, Noise ReductionI. INTRODUCTIONIn imaging science and computer vision, image processingis a technical analysis of the complex aspects of an image,deploying algorithms in which the input is an image, such asa photograph or video frame; the output of image processingmay be either an image or a set of characteristics or parametersrelated to the image. Worldwide the eld of image processingis growing rapidly.
An image is a visual representation of something. In informationtechnology, the term has several usages:i. An image is a picture that has been created or copied andstored in electronic form. An image can be described in termsof vector graphics or raster graphics.
ii. An image is a section of random access memory ( RAM) that has been copied to another memory or storage location.A digital image is a representation of a two-dimensionalimage as a finite set of digital values, called picture elementsor pixels. Pixel values typically represent gray levels, colors,heights, opacities etc. Digitization implies that a digital imageis an approximation of a real scene.Pixel is the basic unit of a digital image which is also knownas picture element. Pixel describes programmable color on acomputer display of each point of an image.
In digital imaging,a pixel, or picture element is a physical point in a rasterimage, or the smallest addressable element in an all pointsaddressable display device.In color image systems, a color istypically represented by three component intensities such asred, green, and blue.An image can be corrupted by various kind of motion, noiseor signal during the process of acquisition. To detect this motionor noise in the picture various method of restoration hasbeen introduced in image processing 7. In general, the noisein the image is defined by the regions which are remarkablydifferent i.e. darker or shiner comparing with the pixel withthe neighboring pixels. To overcome this problem author rstly,focused on the causes of blur or corrupted image, amongvarious method of restoration, emphasizes the neighboringpixel method, developed an algorithm for the restoration ofan image by using distance transformation and mean method.
Image processing is a method which includes some operationsapplying in any image in order to get an enhanced imageor to extract some useful information from it. Image processing7 refers to the analysis and manipulation of a digitized image,especially in order to improve its quality.Digital Image is a two-dimensional function x and y arespatial coordinates. The amplitude of f(x,y) is called intensityor gray level at the point (x,y).Image processing refers to the analysis and manipulation ofa digitized image, especially in order to improve its quality.
The Author emphasizes on solving identification problemsprimarily noticed in forensic medicine, or in the creation ofweather maps from satellite images. The bitmapped graphicsformat images that have been scanned in or shot with digitalcameras – which is used to deal with the reconstruction orenhancement of the uncorrupted image from a noisy one.A good number of filtering schemes such as Wiener filter4,Bilateral filter 12, Bayseian based iterative method 10 aredeveloped over time for noise reduction and image restoration.
The Median filter 2 is one the filtering schemes that arewidely used for noise reduction and also have applications indigital signal processing.The main objective of this paper is to introduce an approachfor image restoration, whose main aim is to make an imagenoise-free. The entire paper is organized in the following sequence.In section-1, Literature review about image restorationhas been proposed. In section-2, the nearest neighbour methodand mean method including algorithm has been introduced.In section-3, the result obtained for the implementation ofalgorithm in MATLAB has been presented with analysis.Finally, the paper concludes with conclusion and references.
II. RELATED WORKImages can be corrupted for various reasons such as motionblur, noise, and camera misfocus. To improve the quality of adegraded image, restoration has to be done. Image restorationis the task of minimizing the degradation in an image i.e. to removethe noise or motion blur. Image restoration assures goodinsights of image when it is subjected to further techniques ofimage processing7.
Image restoration refers to remove motion or distortion fromimage and get an improved quality of that defected image.Images are often degraded during the data acquisition process.The degradation may involve blurring,information loss due tosampling, quantization effects, and various sources of noise.Therefore, the purpose of image restoration is to estimate theoriginal image from the degraded data. Applications of imagerestoration5. are needed in various sector like from medicalimaging, astronomical imaging,to forensic science, etc. Oftenthe benefits of improving image quality to the maximumpossible extent far out weight the cost and complexity of therestoration algorithms involved 1.Lee and Jong-Sen6 described Computational techniqueson 2D image arrays that are developed on the basis ofimages’ local mean and variance to contrast enhancement andnoise filtering.
They rely on nor-recursive algorithms and didnot use any kind of image transform. The proposed methodis obvious in real-time image processing applications withparallel processing.An algorithm in MATLAB which is based on the neighborhoodproperty of a pixel. We focus on a certain iterativeprocess to carry out restoration. One such method described inthis regard is the Mr. JAGADISH H.
PUJAR and Mr.KIRANS. KUNNUR proposed an image restoration Nearest NeighbourhoodMethod 5.E.P. Simoncelli, E.
H. Adelson11 introduces a methodwhich is a classical solution to the noise removal problem isthe Wiener filter, which utilizes the second-order statistics ofthe Fourier decomposition. Authors developed an extensionof the Wiener solution which is a Bayesian estimator thatperforms a “coring” operation, and a simple model of thesubband statistics to develop a semi-blind noise removalalgorithm which relies on a steerable wavelet pyramid.
K. Miyata, N. Tsumura, H Haneishi and Y Miyake 8proposed a Wiener filtering method that can improve thetotal quality of images corrupted by additive noise withoutdegrading the sharpness caused by the noise reduction process.
They observed covariance matrices of the images are estimatedfrom the neighboring pixels which are selected around thecurrent pixel with a color classification technique.Nowak, Robert D proposed a novel wavelet-domain noisereduction procedure that is diversified to variations in signaland to the noise. It is believed that magnetic resonance andmagnitude image data follows a Rician distribution 9.
TheRician noise is troublesome in low Signal to Noise Ratio(SNR) domain to which it creates random fluctuations butreduces noise. In that work, authors studied and proposed amethod for Rician noise removal.Statistical data analysis also helps to remove noise fromtarget images. Garnett, Roman and Huegerich, Timothy andChui, Charles and He, Wenjie introduced image statistic thatidentifies noise pixels with impulse noise of random values ofimage pixels 3. They quantify intensity differentiation fromneighbor pixels, demonstrated how proposed image statisticcan be introduced to remove additive Gaussian noise andcapable to reduce both Gaussian and impulse noises effectivelyfrom the noisy images.
III. PROPOSED WORKDigital Image processing stems from various tasks of practicaltechniques such as Classification, Feature extraction,Multi-scale signal analysis, Pattern recognition and Projectionand these tasks are almost impossible to solve efficiently usinganalog apparatus methods. To restore noisy image many filtershave been proposed like Guided filter, Gaussian filter, Wienerfilter and Median filters are some of the restoration techniques.We have deviced a new approach to restore a RGB image usingNeighboring Piexl with mean method.A noisy RGB image was taken as input and calculated itsnoise in MATLAB. Then we have taken the nearest neighborpixels, 8 for 3×3 matrix and extracted a submatrix of eachpixel.
Calculate the mean of submatrix for each pixel. Then replacethe center pixel with the mean of its neighboring pixels.Using an iterative process each pixel has been considered andtraversed.
After the whole traversal, we have got our desiredoutput image.Then we again calculate the noise of restoredimage and compare it with previous noise.We developed a prototype using Matlab. The entire processhas been represented in Fig. 1. Sample images are collectedfrom Wikipedia images.A.
Image AcquisitionThere are some fundamental steps of image processing. ImageAcquisition is the first step or process of the fundamentalsteps of digital image processing. Image acquisition could beas simple as being given an image that is already in digitalform.
If not we can easily convert it to digital form. Generally,the image acquisition stage involves preprocessing. In ourmethod we take a noisy RGB image as input for restorationand calculated its noise in MATLAB.element m¯ j is the mean value of the N observations of theJth variable:m¯ j =1NXNi=1mi,j j = 1, …, K (1)Thus, the sample mean vector contains the average of theobservations for each variable, and is written as:m¯ =1NXNi=1mi =??????????????m¯ 1.
..m¯ K??????????????(2)E. Pixel ReplacementAbove mentioned, to carry out restoration, we considerthe mean of nearest neighbours of a pixel. In our approach,we consider a 3×3 matrix, total of eight neighbours of eachpixel.The size of the window can be more than 3×3 too.
In the2D grid of picture, each pixel has a certain correlation withits nearest pixels. With the aid of this property, we introduce amethod to replace a noisy pixel by a value which happens tobe the mean of all the nearest neighbors in a filtering windowof 3×3. This ensures a good level of noise reduction fromimage as shown in the results.
IV. RESULT ANALYSISImage Restoration is the operation of taking a corrupted ornoisy image and estimating the clean, original image in whichthe input is an image, such as a photograph or video frame;the output of image processing may be either an image or aset of characteristics or parameters related to the image. Ourapproach was applied on both a noisy RGB image and a noisygrayscale images and obtained much better result comparingother noise removal schemes.Fig. 4: Noisy Image vs Restored Image.In Fig-4, the input image is a noisy image which has somedegree of noise density in it and the restored image eachcorrupt pixel is replaced by the mean value of its neighbours.
We tested on several standard test images and the above resultwas obtained for the image which is corrupted by noise. Asa result of this, the above result was obtained for mean ofneighboring pixel i.e. for a given pixel at (i,j), all the eightneighbours of it are taken into account for restoring a pixel at(i,j).
Fig. 5: Noise after applying different filters for RGB imageThe bar graphs comparison of resulted noise for an RGBimage after applying different image restoration techniquesi.e. Guided filter, Gaussian filter. It can be clearly seenthat the amount of noise reduced by proposed method issignificantly high. In original image, amount of noise calculatedapproximately 4×10?3 using available noise estimationroutine. Guided filter, Gaussian filter remove the noise atapproximately 2×10?3and proposed method reduce the noisesignificantly to 1×10?3, that is, proposed method reduces noiseby 79% of original noise.
RestorationMethodImageFormatOriginalNoiseResultedNoiseReductionRateGuided filter RGB 3.9×10?3 2.3×10?3 41%Gaussian filter RGB 3.9×10?3 1.9×10?3 51%Wiener filter RGB N/A N/A N/AMedian filter RGB N/A N/A N/AProposedMethod*RGB 3.9×10?3 0.8×10?3 79%TABLE I: Noise level for different Filters using RGB image.We could not calculate noise level of Wiener filter andMedian filter as mentioned in table by N/A.
Fig. 6: Noise after applying different filters for grayscale imageWe experimented different filters like Guided filter, Gaussianfilter, Wiener filter, Median filters for grayscale images aswell and proposed method outperformed the mentioned noiseremoval techniques. Initial noise level observed in originalimage is approximately 0.03 and mentioned filters the removenoise level whereas best output obtained by Median filter.However, the proposed method reduces the noise level to0.004.
The result obtained is shown in both Table-1 and Fig.6, that is proposed method reduces 85% of the noise fromoriginal image.RestorationMethodImageFormatOriginalNoiseResultedNoiseReductionRateGuided filter RGB 0.027 0.
017 37%Gaussian filter RGB 0.027 0.013 51%Wiener filter RGB 0.027 0.008 70%Median filter RGB 0.
027 0.006 77%ProposedMethod*RGB 0.027 0.004 85%TABLE II: Noise level for different Filters using Grayscaleimage.V. CONCLUSION AND FUTURE WORKIn this work, we mainly focused nearest neighborhood pixeland mean method.
The algorithm combining these methods areable to restore the image to a significant level. We take a submatrix of 3/3 and calculate the mean of the sub matrix. Thenreplace the center pixel with the mean of neighboring pixels.The restored image obtained after performing these steps areof much better.However, using proposed method pixels at the boundarycannot be considered as the center pixels. As a result, afterapplying the proposed technique the given image holds noiseor blur at the edges. Therefore, boundaries noise cannot bereduced as desired an in future that could be focused on.
For the further enhancement of image, various featuredetection method like edge detection methods like sobel edgedetection technique or various filtering can be applied and usedon the basis of efficiency to remove the noise completely atthe edges and boundaries of the pictures.REFERENCES1 Mark R Banham and Aggelos K Katsaggelos. “Digitalimage restoration”.
In: IEEE signal processing magazine14.2 (1997), pp. 24–41.2 Tao Chen, Kai-Kuang Ma, and Li-Hui Chen. “Tri-statemedian filter for image denoising”. In: IEEE Transactionson Image processing 8.
12 (1999), pp. 1834–1838.3 Roman Garnett et al. “A universal noise removal algorithmwith an impulse detector”.
In: IEEE Transactionson image processing 14.11 (2005), pp. 1747–1754.4 Allen D Hillery and Roland T Chin. “Iterative Wienerfilters for image restoration”. In: IEEE Transactions onSignal Processing 39.
8 (1991), pp. 1892–1899.5 H PUJAR JAGADISH et al. “A novel approach forimage restoration via nearest neighbour method”. In:Journal of Theoretical and Applied Information Technology14.2 (2010).6 Jong-Sen Lee. “Digital image enhancement and noisefiltering by use of local statistics”.
In: IEEE transactionson pattern analysis and machine intelligence 2 (1980),pp. 165–168.7 Robert M Lougheed. Neighborhood image processingstage for implementing filtering operations. US Patent4,541,116. Sept. 1985.
8 Kimiyoshi Miyata et al. “Restoration of Noisy Imagesusing Wiener Filters Designed in Color Space”. In: ISAND TS PICS CONFERENCE. SOCIETY FOR IMAGINGSCIENCE & TECHNOLOGY. 2000, pp.
301–306.9 Robert D Nowak. “Wavelet-based Rician noise removalfor magnetic resonance imaging”. In: IEEE Transactionson Image Processing 8.10 (1999), pp. 1408–1419.10 William Hadley Richardson. “Bayesian-based iterativemethod of image restoration”.
In: JOSA 62.1 (1972),pp. 55–59.11 Eero P Simoncelli and Edward H Adelson. “Noiseremoval via Bayesian wavelet coring”.
In: Image Processing,1996. Proceedings., International Conferenceon.
Vol. 1. IEEE. 1996, pp. 379–382.
12 Buyue Zhang and Jan P Allebach. “Adaptive bilateralfilter for sharpness enhancement and noise removal”.In: IEEE transactions on Image Processing 17.5 (2008),pp. 664–678.