A Detail-Preserving Type-2FuzzyLogicFilter forImpulse Noise RemovalfromDigital Images M.Tiilin YILDIRIM, Student Member, IEEE,AlperBASTURK,Member, IEEE,and M.EminYUKSEL,Senior Member, IEEE
Alper Ba · 2007
A novelfiltering operator basedontype-2 fuzzy logic techniques isproposed fordetail preserving restoration ofimpulse noisecorrupted images. Theperformance ofthe proposed operator istested fordifferent test images corrupted atvarious noise densities andalsocompared withrepresentative conventional aswellasstate-of-the-art impulse noise removal operators fromtheliterature. Experimental results showthat theproposed operator exhibits superior performance overthe competing operators andiscapable ofefficiently suppressing the noise intheimagewhile atthesametimeeffectively preserving theuseful information intheimage. I.INTRODUCTION Digital images areoften contaminated byimpulse noise during imageacquisition and/or transmission duetoanum- berofimperfections encountered inimage sensors andcom- munication channels. Inmostimage processing applications, itisveryimportant toremovetheimpulse noise fromthe imagebecause theperformances ofsubsequent imagepro- cessing tasks, suchasedgedetection, imagesegmentation, object recognition, etc., areseverely degraded bynoise. A large number ofmethods havebeenproposed toremove impulse noise fromdigital images. Themajority ofthese methods arebasedonorder statistics filters, whichutilize therankorder information ofthepixels contained inagiven filtering window. Thestandard median filter (1)attempts to removeimpulse noise bychanging thecenter pixel ofthe filtering windowwiththemedian ofthepixels within the window. Thisapproach provides areasonable noise removal performance butremoves thin lines andblurs image details evenatlownoisedensities. Theweighted median filter andthecenter-weighted median filter (2)-(4), whichare modified median filters giving moreweight tocertain pixels inthefiltering window, havebeenproposed toavoid the inherent drawbacks ofthestandard median filter. These filters demonstrate better performance inpreserving image details than themedian filter attheexpense ofreduced noise removal performance.