S.W.Franklin,S.E.Rajan/AppliedSoftComputing22(2014)94–100

97
Fig.2.Preprocessedimagesobtainedfromoriginalcolorfundusimagesafterapplyingmean ltering.

Fig.3.PreprocessedimagesafterapplyingGaussian lters.

Fig.4.AnexampleforGaborfeaturesextractedatdifferentorientations.
aspertheGaborcoef cientsattained,theoriginalimagesarecon-volved.Next,themomentinvariants-basedfeaturesareextracted.
Nowthesamplesaretakenfromvesselandnonvesselregionsandtheyaretrainedusingneuralnetworks.Thefeaturevectorsextractedfrompreprocessedretinalimagesaregivenasinputtothisneuralnetwork.Thistechniqueisbasedonpixelclassi cationandtheseclassi cationresultsarethresholdedtoclassifyeachpixelasvesselornonvessel.
Neuralnetworktraining
Eachpixelofaretinalimageisclassi edasvesselornonves-sel,usingneuralnetworks.Thecomputationalstructureofneural
networksandthehumanvisualsystemhasseveralsimilarities.Inordertoclassifyeachpixeloftheretinalimageasvesselornot,amulti-layerperceptronneuralnetworkisused[27,28].Itisamulti-layer,feedforwardnetworkusingextendgradient-descentbaseddelta-learningrule,commonlyknownasbackpropagationrule.Backpropagationisasystematicmethodfortrainingmulti-layerarti cialneuralnetwork.Accordingtothebasicde nitionoffeedforwardnetwork,aneuroninthehiddenlayerwillreceiveinputsfromallneuronsinthepreviouslayerandmustfeedallneuronsinthenextlayer.Backpropagationprovidesanef cienttechniquetochangetheweightsinafeedforwardnetworkwithdifferentiableactivationfunctionunits,andtolearnatrainingsetofinput–outputexamples.Ifthenumberoflayersinthefeedforwardnetworkis
ICPR06_A new approach to automated retinal vessel segmentation using ...Gabor filters and scale production, and a threshold probing technique ...
热门文档