A DCTBasedFiltering ofBiornedical Irnages
Pradeep K. Gupta · 2006
Imagefiltering techniques havenumerous po- sumingaspecific modelfortheobservations, together with tential applications inbiomedical imagingandimagepro-a priormodelfortheunknownimage.The estimates sat- cessing. Thedesignoffilters largely dependsonthea-priori knowledge aboutthetypeofnoisecorrupting theimageisfy well-known optimality properties underthespecified andimagefeatures. Thismakesthestandard filters tobe model.Unfortunately, thetruemodelisnotavailable in application andimagespecific. Themostpopular filters practice. Itthenbecomes essential toevaluate theperfor- suchasaverage, Gaussianand Wienerreducenoisyarti-manceoftheestimator underdepartures fromtheassumed factsby smoothing.However,thisoperation normallyre- sultsinsmoothing oftheedgesaswell.On theotherhand, model.Wavelettransform domainfiltering methodshave sharpening filters enhancethehighfrequency details mak- becomeverypopular inlastdecade.Thereasonbehind ingtheimagenon-smooth. An integrated general approachthesuccess ofwavelettransform isthatitprovides good todesignfilters basedondiscrete cosinetransform(DCT)is localization inbothspatial andspectral domains, allowing proposed inthisstudyforoptimalmedicalimagefiltering. Thisalgorithm exploits thebetter energycompaction prop- noiseremovalandedgepreservation/ enhancement. Nu- ertyofDCT andre-arrange thesecoefficients ina waveletmerouswavelet techniques havebeendeveloped fordenois- mannertogetthebetterenergyclustering atdesiredspa- tial locations. Thisalgorithm performs optimal smoothingingmonochromeimages(16)(18) ofthenoisyimagebypreserving highandlowfrequency DCT baseddenoising techniques arealsopopular forim- features. Evaluation results showthattheproposed filter is agerestoration (20)- (23).A feature preserving noisere- robustundervarious noisedistributions. movalalgorithm basedonDCT andthea-prior knowledge ofpixel typeisdiscussed in(19). Songetal.haveproposed