BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//132.216.98.100//NONSGML kigkonsult.se iCalcreator 2.20.4//
BEGIN:VEVENT
UID:20260808T153857EDT-67474s1AT3@132.216.98.100
DTSTAMP:20260808T193857Z
DESCRIPTION:This Wednesday\, Dr Nadja Schinkel-Bielefeld\,\nfrom the Depart
 ment of Brain and Cognitive Sciences\,\nMIT will give a presentation title
 d: ' Connections\nUnderlying Human Contour Integration are Adapted to Natu
 ral Image\nStatistics' (abstract below). Dr Schinkel-Beilefeld\nis an invi
 ted speaker\, who is being hosted by Dr Kathy Mullen\nWhere: Royal Victori
 a Hospital\, Room H4-14\nWhen: Wednesday\, 18th of February 2009\, 1PM\nAb
 stract:Humans are very efficient in detecting even jittered or\npartly occ
 luded contours. But which mechanisms are capable of\nexplaining this abili
 ty? A large class of contour integration\nmodels employs a so-called assoc
 iation field (AF)\, which reflects\nthe connection structure between neuro
 nal units representing\ndifferent edge elements\, for linking localized ed
 ges to a global\ncontour. Here we investigate systematically how such an A
 F should\nbe constructed to explain psychophysical contour detection data\
 ,\ncombining ideas from theory\, experiment and natural image\nstatistics.
  We tested AFs of different symmetry and geometry in a\nBayesian contour i
 ntegration model\, comparing it to human contour\ndetection. Our psychophy
 sical experiments revealed that\ncorrelations between human decisions are 
 much higher than expected\nfrom their detection performances alone. This s
 uggests that\ndifferent humans make identical errors\, voting for the same
 \nillusory contours in the stimulus. Hence we require our contour\nintegra
 tion model not only to reach human performance\, but also to\nreproduce th
 ese correlations and their characteristic time course.\nFrom all the AF ch
 oices we explored\, a connection structure\nextracted from natural image s
 tatistics explains human contour\nintegration best. From this we conclude 
 that the human visual\nsystem is optimized for recognizing contours in nat
 ural images.\n
DTSTART:20090218T180000Z
DTEND:20090218T190000Z
SUMMARY:MVR Journal Club
URL:/channels/event/mvr-journal-club-104577
END:VEVENT
END:VCALENDAR