Theautomaticdetectionoffacesisaveryimportantproblem.Theeffectivenessofbiometricauthenticationbasedonfacemainlydependsonthemethodusedtolocatethefaceintheimage.Thispaperpresentsahybridsystemforfacesdetectioninunconstrainedcasesinwhichtheillumination,pose,occlusion,andsizeofthefaceareuncontrolled.Todothis,thenewmethodofdetectionproposedinthispaperisbasedprimarilyonatechniqueofautomaticlearningbyusingthedecisionofthreeneuralnetworks,atechniqueofenergycompactionbyusingthediscretecosinetransform,andatechniqueofsegmentationbythecolorofhumanskin.Awholeofpictures(facesandnofaces)aretransformedtovectorsofdatawhichwillbeusedforlearningtheneuralnetworkstoseparatebetweenthetwoclasses.Discretecosinetransformisusedtoreducethedimensionofthevectors,toeliminatetheredundanciesofinformation,andtostoreonlytheusefulinformationinaminimumnumberofcoefficientswhilethesegmentationisusedtoreducethespaceofresearchintheimage.Theexperimentalresultshaveshownthatthishybridizationofmethodswillgiveaverysignificantimprovementoftherateoftherecognition,qualityofdetection,andthetimeofexecution.
