視覺(jué)信息處理圖像成分組合和分割的神經(jīng)機(jī)制

StateKeyLaboratoryofCognitiveNeuroscienceandLearning ernInstituteforBrainResearchBeijingNormal 年月“Thesensationsarousedbylightinthenervousmechanismofvisionenable existence,formandpositionexternalobjects.Theseideascalledvisualperceptions.”

(1821–1894)

TreatiseonPhysiologicalOptics, 2①Visualperceptionisdependentonstimulus②Visualperceptionisdependentonselective③Visualperceptionisdependentonpast③VisualperceptionisdependentonpastWangetal.DynamicandadaptivevisualVisualperceptionisdependentonstimulusandbehavioralcontextandshapedbypastexperience.Whataretheneuralmechanismsunderlyingsuchflexibleperceptualcapabilities?e

cortex,V1)

視網(wǎng)膜神經(jīng)節(jié)細(xì)胞（retinalganglioncell）的感受野（receptivefield）特性452452DoGDoGOptimalOrientationOptimalOrientationMeanMeanresponse(0

-45

OrientationtuningofV1neurons(Hubel&Wiesel,FormingasimplecellinV1fromLGNcells(Hubel&Wiesel,=X=Gabor HierarchicaltheoryinvisualHubel&Wiesel,Max(1880- andsegmentationinvisual B TheGestaltlawofContourNeuralcircuitryintheprimaryvisualcortex

contourintegrationContextualinfluencesonneuronalresponsesin Firingrate(Firingrate(

Li&Li,VisionResearchContourintegrationinWhichofWhichofthetwopatternscontainsaContourintegrationinV1:correlationwithperceptualLietal.,NeuronHitHit FalsealarmrateSignaldetectionContourintegrationinV1:choice NumberofcollinearLietal.,NeuronContourintegrationinV1:attentionalLietal.,NeuronMorethanhard-wiredprocessingwithgainInfluencesoftop-downanticipationonV1Samplecue:380 Detection:48-191McManus,Li&Gilbert,PNASInfluencesoftop-downanticipationonV1McManus,Li&Gilbert,PNASPerceptualContour97531CentralContour97531Central97531es/s)

LearningLearningsticityin Lietal.,NeuronLearningLearningsticityinAnesthetizedAnesthetized0 Lietal.,Neuron ctionsbetweenlowerandhighervisual ctionsbetweenlowerandhighervisualV1 UtahChenetal.,NeuronContour-inducedresponsesinV1and IncrementalcontourintegrationbetweenV1andContourintegrationstartsinitiallyinThefastfeedforwardprocessisfollowedbyadelayedprocessthatengagesV1neuronsonthecontourandonthebackground.AfterV1isengagedthecontoursignalsinbothareascontinuetobuildupandrapidlyreachthe IncrementalcontourintegrationbetweenV1andGrangercausalityFortwosimultaneouslymeasuredtimeseriesXandY,XcanbecausaltoYifwecanpredictYbypastknowledgeof

-300to9595to950Amodelwithtop-downgainPiech,Li,Reeke&Gilbert,PNASAmodelwithtop-downgainPiech,Li,Reeke&Gilbert,PNASAmodelwithtop-downgainAcomparisonbetweensimulationandrealPiech,Li,Reeke&Gilbert,PNASAmodelwithtop-downgainPiech,Li,Reeke&Gilbert,PNASSurfacesegmentationbyvisualcorticalPoortetal.,NeuronVisualinformationprocessinginthebrainisdynamicallymodifiedbystimulusandbehavioralcontextandshapedbytraining.Parsingvisualimagesrequiresinctionsamongcorticalneuronsacrossmultipleareas.Fastfeedforwardinputscreateacoarsetemteinhighervisualcortex,whichgatestheprocessinginearlycorticalareasatfinerscalesthroughfeedbackmodulations.ThiscountercurrentprocessingschemeisinaccordancewiththeGestalttheory,whichpositsthatinterpretationsoflocalimagecomponentsareconstrainedbytheglobalorganizationofvisualscenes.Repeatedlyperformingthesameperceptualtask,andtherefore,repetitivelyinvokingtop-downinfluencescanpotentiateadaptivechangesinthevisualcortex,allowingforamoreefficientreadoutoftask-relevantinformation.Chen,M.,Yan,Y.,Gong,X.,Gilbert,CharlesD.,Liang,H.,andLi,W.(2014).Incrementalintegrationofglobalcontoursthroughinterybetweenvisualcorticalareas.Neuron82,Gilbert,C.D.,andLi,W.(2013).Top-downinfluencesonvisualprocessing.NatureReviewsNeuroscience14,350-363.Li,W.,Piech,V.,andGilbert,C.D.(2006).Contoursaliencyinprimaryvisualcortex.Neuron50,951-962.Li,W.,Piech,V.,andGilbert,C.D.(2008).Learningtolinkvisualcontours.Neuron57,442-McManus,J.N.J.,Li,W.,andGilbert,C.D.(2011).Adaptiveshapeprocessinginprimaryvisualcortex.Proc.Natl.Acad.Sci.U.S.A.108,9739-9746.Pi?ch,V.,Li,W.,Reeke,G.N.,andGilbert,C.D.(2013).Networkmodeloftop-downinfluencesonlocalgainandcontextualinctionsinvisualcortex.Proc.