Visual crowding illustrates the inadequacy of local vs. global and feedforward vs. feedback distinctions in modeling visual perception

Abstract

Experimentaliststendtoclassifymodelsofvisualperceptionasbeingeitherlocalorglobal,andinvolvingeitherfeedforwardorfeedbackprocessing.Wearguethatthesedistinctionsarenotashelpfulastheymightappear,andweillustratetheseissuesbyanalyzingmodelsofvisualcrowdingasanexample.Recentstudieshavearguedthatcrowdingcannotbeexplainedbypurelylocalprocessing,butthatinstead,globalfactorssuchasperceptualgroupingarecrucial.Theoriesofperceptualgrouping,inturn,ofteninvokefeedbackconnectionsasawaytoaccountfortheirglobalproperties.Weexaminedthreetypesofcrowdingmodelsthatarerepresentativeofglobalprocessingmodels,andtwoofwhichemployfeedbackprocessing:amodelbasedonFourierfiltering,afeedbackneuralnetwork,andaspecificfeedbackneuralarchitecturethatexplicitlymodelsperceptualgrouping.Simulationsdemonstratethatcrucialempiricalfindingsarenotaccountedforbyanyofthemodels.Weconcludethatempiricalinvestigationsthatrejectalocalorfeedforwardarchitectureofferalmostnoconstraintsformodelconstruction,asthereareanuncountablenumberofglobalandfeedbacksystems.Weproposethattheidentificationofasystemasbeinglocalorglobalandfeedforwardorfeedbackislessimportantthantheidentificationofasystem’scomputationaldetails.Onlythelatterinformationcanprovideconstraintsonmodeldevelopmentandpromotequantitativeexplanationsofcomplexphenomena.