Object recognition 3Visual cognition theoriesFeature detection in object recognitionDenis PelliPsychology and Neural Science, NYUDecember 10, 2007Outline1. Critical comments on existing theories, in visual cognition, for how peoplerecognize objects.2. FeaturesOutline1. Critical comments on existing theories, in visual cognition, for how peoplerecognize objects.Treisman & Kanwisher 1998 Perceiving visually presented objects: recognition, awareness, andmodularity. Current Opinion in Neurobiology.2. FeaturesTreisman & Kanwisher 1998 Perceiving visually presented objects: recognition, awareness, andmodularity. Current Opinion in Neurobiology.How do we recognize objects? What are the theories? The centralcontroversies in the field have been dichotomies:Parts vs. wholeStructural vs. viewpoint-dependentConscious vs. unconsciousPerception vs. actionDorsal vs. ventralLots of papers have accumulated evidence bearing on these issues, showingdifferences along these scales between tasks. In some cases patients revealdissociations. These dimensions are real; there are data to be explained.However, while initial positions were at either ends of the dichotomies,everyone has since drifted to moderate views that allow for intermediatepositions or combinations of both extremes. As a consequence, no one iswrong. Since no one is wrong, one may well wonder whether the intermediatepositions that embrace the whole gamut are testable scientific theories. Canthey be refuted?Except for part/whole in face recognition, these theories haven’t helped much inexplaining everyday object recognition.Treisman & Kanwisher 1998 Perceiving visually presented objects: recognition, awareness, andmodularity. Current Opinion in Neurobiology.Priming, matching, and repetition blindness are all object-specific, yet invariantacross views.Object files (Pylyshyn, multiple object tracking)Pylyshyn, multiple object trackingTreisman & Kanwisher 1998 Perceiving visually presented objects: recognition, awareness, andmodularity. Current Opinion in Neurobiology.“Wolfe (1998) has collected surprising evidence that previously attended objecttokens revert to a similar unstructured state once attention is withdrawn,concluding that, ‘Vision exists in the present tense. It remembers nothing.’”Treisman & Kanwisher 1998 Perceiving visually presented objects: recognition, awareness, andmodularity. Current Opinion in Neurobiology.ModularityModularity vs. consciousnessThere is much evidence of modularity in the brain, in which someareas seem to know things that other areas don’t. Viewing thebrain as a machine, this is an old familiar result from thenineteenth century, and a perfectly reasonable way to build abrain. However, if we are talking about the human brain, then themodularity challenges our intuitions about consciousness.Ned Block distinguishes between phenomenal consciousness andaccess consciousness. Phenomenal consciousness consists ofsubjective experience and feelings. Access consciousnessconsists of that information globally available in the cognitivesystem for the purposes of reasoning, speech and high-levelaction control. The key word is “globally”. Our intuition is thatinformation should never be inconsistent among different parts ofthe same brain, or different aspects of our behavior.Outline1. Critical comments on existing theories, in visual cognition, for how peoplerecognize objects.2. FeaturesSame features for letter identification as for grating detection.Features do not scale with letter size.Features are detected independently.Solomon & Pelli (1994) NatureSolomon & Pelli (1994) NatureGainc/letter0.150.62.49.7Majaj, Liang, Martelli, Berger & Pelli (2003) Vision Sciences SocietyMajaj, Pelli, Kurshan, & Palomares (2002) Vision ResearchMajaj, Pelli, Kurshan, & Palomares (2002) Vision ResearchMajaj, Pelli, Kurshan, & Palomares (2002) Vision ResearchMajaj, Pelli, Kurshan, & Palomares (2002) Vision ResearchMajaj, Pelli, Kurshan, & Palomares (2002) Vision ResearchMajaj, Pelli, Kurshan, & Palomares (2002) Vision ResearchMajaj, Pelli, Kurshan, & Palomares (2002) Vision ResearchMajaj, Pelli, Kurshan, & Palomares (2002) Vision ResearchRobson & Graham (1981) Vision ResearchEnglish (Bookman) abcdefghijklmnopqrstuvwxyzEnglish (uppercase Bookman) ABCDEFGHIJKLMNOPQRSTUVWXYZEnglish (bold Bookman) abcdefghijklmnopqrstuvwxyzEnglish (Courier) abcdefghijklmnopqrstuvwxyzEnglish (Helvetica) a b c d e f g h i j k l m n o p q r s t u v w x y zEnglish (Künstler) A B C D E F G H I J K L M N O P Q R S T U … ZSloan CDHKNORSVZArabic x , iK LG: R T Y U34Q W A S.v5 6OP { E J F hArmenian abgd;xhejvil.`kf][ym\n,ocp=®swtrz'qChineseabcdefghijklmnopqrstuvwxyzDevanagari a b c d E f g h D j k l m n Z p q r s t A v W x y zHebrew tcrqxp[snmlkyfjzwhdgba2! 3 Checkers abcdefghijklmnopqrstuvwxyz4! 4 Checkersabcdefghijklmnopqrstuvwxyz3-letter words all and any are but can for had has her him his its … you5-letter words about after being could first great house might never … years0.010.1110 100Figure 4aEfficiencyComplexity daaKafmanAaawasawhichAaPelli, Burns, Farell, & Moore (2005) Vision ResearchEfficiency perimeter2“ink” areaComplexityPelli, Farell, & Moore (2003) Nature0.010.11 10EfficiencyWord!length!(letters)bPelli, Farell, & Moore (2003) Nature1010010001 10Energy (re noise)Word!length!(letters)aIdealHumanPelli, Farell, & Moore (2003) NatureOf course you can read letter by letter, but can you read word by word? Pelli, Farell, & Moore (2003) NatureOutline1. Critical comments on existing theories, in visual cognition, for how peoplerecognize objects.Treisman & Kanwisher 1998 Perceiving visually presented objects: recognition, awareness, andmodularity. Current Opinion in Neurobiology.2. FeaturesSame features for letter identification as for grating detection.Features do not scale with letter size.Features are detected