IJCST Vo L . 3, ISS Ue 2, A - Ju N E ISSN : 0976-8491 .

IJCST Vol. 3, Issue 2, April - June 2012ISSN : 0976-8491 (Online) ISSN : 2229-4333 (Print)New Fuzzy Logic Based Filter for Reducing NoisesFrom Images11,2Sanyam Anand, 2Navjeet KaurDept. of CSE, Lovely Professional University, phagwara, Jalandhar, Punjab, IndiaAbstractThe noise present in the image can coat and lessen the visibilityof certain characteristics of the objects present within the image.Thus making it unfit for further enhancements. There has beena research on the denoising filters for very long to remove thenoise from images. Here, in this paper we propose the enhancedfuzzy classical filter which removes more than one type of noisesfrom images. Traditional mean and median filters are combined toremove more types of noises. Performance of the filter is comparedwith other filters and the proposed technique has outperformedexisting techniques.KeywordsNoise Reduction, Mixed Noise Removal, Fuzzy ClassicalFilters.I. IntroductionImage Processing is a practice to make better raw imagescaptured from cameras/sensors positioned on satellites, spaceand aircrafts or pictures taken in normal routine life used in arange of applications. While acquiring the image from the supplylike sensor, digital camera etc. there may occur some instability orwe can describe it as noise. Main contributor to noise occurrenceincludes bad weather surroundings or some other interventionwhile transmitting the image. The outcome is infected imagethat require to be pre-processed to lessen or remove the noise.Presence of noise makes an image the undesirable one, which isnot suitable for further image enhancement tasks like zooming,object detection, pattern recognition etc. So, they have to beimproved before other analysis or enhancement techniques canbe performed on them. There are different types of noises eachhaving different statistical properties [19]. A variety of techniquesare used to eliminate these noises. However a universal “best”approach has yet to be found.This paper proposes the new technique based on fuzzy logic whichlessens the noise from grey scale images. It is a fuzzy classicalmethod. It means classical techniques are enhanced by mergingthe fuzzy logic rules in them. The focus is put on removal of morethan one type of noise, it means that the presented techniques willable to remove the two or more types of noises from differentimages or from same image if exists.The structure of the paper is organized as follows: In Section II,related work done in this area is presented. Literature survey isdone for this purpose. In Section III, proposed work is explainedalong with pseudo code. Section IV, presents several experimentalresults. These results are discussed in detail, and are comparedto those obtained by other filters. Final conclusions are drawn inSection V.II. Related WorkMany researchers have done research in filters related to removingnoises from images. But very little work on the filters whichremoves more than one types of noise. Some study on relatedwork is done. Latest work done in this field is proposing the fuzzylogic based filter [1], which tries to remove the Gaussian and322International Journal of Computer Science And Technologyimpulse noise from images. It proposes the fuzziness in form ofmembership function. [1] is also taken as the base of our filter. Inanother study [2], the two step method is proposed to detect andremove the noise from images. Further, the method is enhanced[9] to color images. The proposed filter is the fuzzy two-step colorfilter; The fuzzy detection method is based on the calculationof fuzzy gradient values and on fuzzy reasoning. This phasedetermines three separate membership functions that are passedto the filtering step. The technique [3], which detect noisy pixelsusing fuzzy reasoning with lowest uncertainty, and in replacenoisy pixels with a heuristic median filter. Another proposeddenoising technique [4] is based on cascaded median and waveletsfilter. The enhanced edge map generation using multi-resolutioncoherence measurement [6] and adaptive wavelet based-specklenoise reduction is performed using the enhanced edge map. Theedge map obtained from coherence measurement has limitationsuch as noisy or discontinued edge detection. This proposed [6]method overcome this limitation. The nonlinear filtering technique[7], contains two separated steps, an impulse noise detection stepand a reduction step that preserves edge sharpness. Gaussian noiseis considered in noise reduction [8]. The non linear function s usedwhich represents fuzzy rules, considering the uncertainty of signal.Signal adaptive median filtering algorithm for removal of impulsenoise is proposed in [10]. Noise candidates are first selected usingthe homogeneity level, and then a refining process follows toeliminate false detections. New algorithm is proposed [11], whichuses fuzzy gradient method with new fuzzy rules to distinguish thenoise pixels and edge pixels. [13], Presents the technique which isrecursive fuzzy switching median filter which is an extension tothe classical switching median filter by employing fuzzy inferencemechanism. By performing fuzzy reasoning at two different levels,the operator [14], is able to effectively cancel noise pulses withoutdegrading the quality of fine details and textures by using the newfuzzy rule base and inference mechanism. A novel fuzzy filter isproposed for removing mixed noise [15], in order to remove mixednoise efficiently, fuzzy rules are set by using multiple differencevalues between arbitrary two pixels in a filter window. Fuzzyfilter [16], which removes the additive noise. Fuzzy rules arefired to consider every direction around the processed pixel andderivatives are computed and fuzzy derivatives to perform fuzzysmoothing by weighting the contributions of neighboring pixelvalues. [18] Proposed the new fuzzy based algorithm for removingimpulse noise while preserving the edge sharpness of the image.It involves three steps: i)define fuzzy sets, ii) construct a set ofIF-THEN rules, and iii) construct the filter based on the set ofrules.III. Proposed WorkThe proposed method will provide the efficient filter whichremoves the noises from grey scale image and work well thanother existing techniques. To remove more than one noise fromimage, basic traditional that is classical filters, mean filter andmedian filter are combined. The proposed method incorporate thefuzzy logic techniques along with the traditional linear and nonlinear filters which makes it versatile and hence able to reducew w w. i j c s t. c o m

IJCST Vol. 3, Issue 2, April - June 2012ISSN : 0976-8491 (Online) ISSN : 2229-4333 (Print)more than one type of noise.A. AlgorithmThe proposed algorithm is described as follows.1. Take the input image.The input image should be a grey scale image. As we are proposinga new algorithm, so it is first applied on the grey scale images.2. Do convolution using mean filter.(1)Here,is the output image achieved by applying the aboveequation i.e. (1) on the input image. It is effective toreduce gaussian noise from images.3. Convolute using median filter(2)Here,is the output obtained by applying the medianformula on the input image i.e. Median calculates themid value of the given input. Operator med is used for this purpose.It is effective to process impulse noise.4. Use the results from step 2 and 3 and make fuzzy rules. Fivemembership functions are used which all are of triangular typessince it give optimized performance among all. Mamdani modelis used in making FIS. Total 25 rules are made which are used tocalculate the new value of the pixel under processing. These areshown as below using the screenshot.Fig. 2: FIS Editor Showing Output Getting from 2 Inputs6. Defuzzification (COA): After getting the fuzzy result, it isconverted back to the crisp one. For this purpose centre of area(COA) method is used(3)7. Noise free output image: Finally the noise free image is obtainedby using steps 2 to 6.Fig. 1: FIS Rules Editor (25 rules)5. Get fuzzy output using FIS.Fig. 3: Algorithm DesignIV. Experimental AnalysisPerformance of both new proposed algorithms are calculated andcompared with the one presented in [1]. The comparison is madeon the basis of parameters PSNR, MSE and execution time. It isdone for three types of noises:- impulse noise, Gaussian noisew w w. i j c s t. c o mInternational Journal of Computer Science And Technology323

IJCST Vol. 3, Issue 2, April - June 2012ISSN : 0976-8491 (Online) ISSN : 2229-4333 (Print)and speckle noise. The results are shown in Table 1, 2 and 3respectively.Table 1 : Performance of Filters with Impulse NoiseFilter typeimpulseNoiseratio(%)PSNRTime(in sec)MSEFuzzy .223897221.1389Table 1, shows the comparison of three filters with varied degreeof impulse noise ratio. Fig. 4, is the original Rice image, availablein MatLab. Fig. 5, is 5% impulse noise picture on which thesefilters are executed. The resultant figsures are fig. 6 and fig. 7.Following graph1 also shows the reduction in error obtained byproposed method.Fig. 6: Fuzzy Filter [1]Fig. 7: Proposed FilterTable 2: Performance of Filters with Gaussian NoiseFilter typeFuzzy filter[1]Graph 1: MSE vs NoiseProposedfilterGaussianNoise ratio(Variance)PSNRTime(in 22.95040.225177332.2323Table 2, shows the comparison of filters with varied degreevariance of Gaussian noise ratio. Fig. 4, is the original Rice image,available in MatLab. Fig. 8, is .01 Gaussian noise picture on whichthese three filters are executed. The resultant figures are fig. 9and fig. 10. Following graph 2 also shows the reduction in errorobtained by proposed method.Fig. 4: Original ImageGraph 2: MSE vs NoiseFig. 5: 5% Impulse Noise324International Journal of Computer Science And Technologyw w w. i j c s t. c o m

IJCST Vol. 3, Issue 2, April - June 2012ISSN : 0976-8491 (Online) ISSN : 2229-4333 (Print)Fig. 8: Gaussian Noise (v 0.01)Graph 3: MSE vs NoiseFig. 9: Fuzzy Filter [1]Fig. 11: 5% Speckle NoiseFig. 10: Proposed FilterTable 3: Performance of Filters with Speckle NoiseFilter typeFuzzy 0PSNRTime(in 7456.7372101.9536125.0531187.0388290.3286Table 3, shows the comparison of three filters with varied degreeof speckle noise ratio. Fig. 4, is the original Rice image, availablein MatLab. Fig. 11, 5% is speckle noise picture on which thesethree filters are executed. The resultant figures are fig. 12 and fig.12. Following graph 3 also shows the reduction in error obtainedby proposed method.w w w. i j c s t. c o mFig. 12: Fuzzy Filter [1]Fig. 13: Proposed FilterV. ConclusionIn this paper, we proposed techniques to deal with one of the mostresearched problems of digital imagery, i.e. noise. We developtwo techniques based on fuzzy logic. The analysis, of all theobtained experimental results, demonstrates that proposed filteroutperforms for denoising all of the above mentioned images. Thefuzzy filter [1] able to reduce the gaussian noise efficiently butis not good for impulse noise. The statistical experiments showsthat proposed filter not only work well for the gaussian noise butit also proves to be good for impulse noise. The experiments arealso conducted on speckle noise. Here also the proposed filterproves to work well. PSNR is increased using proposed filter whichInternational Journal of Computer Science And Technology325