Please enable JavaScript.
Coggle requires JavaScript to display documents.
Objects and Faces (Object recognition- Classification:viewpoint…
Objects and Faces
Object recognition- Classification:viewpoint generalisation e.g. recognising an object from any viewpoint e.g. viewpoint A, B, Novel viewpoint
one of most difficult tasks visual system has to perform, clear dichotomy, computer scientists been trying to solve for decades
How do we do it? image based models- specific views are stored and recognition performance is somehow based on generalisations from these, they encode structured templates of viewpoint- dependent representations- Riesenhuber & Poggio 2000
How do we do it? structural description models- information about the 3D structure of an object is extracted from a single view, object parts are represented independently of their spatial configuration and viewpoint- David Marr 1982
-
Geons are defined by variations in a small number of basic parameters called non accidental properties - NAP
NAP properties: Curvilinearity- Parallelism, Contermination, Symmetry, Collinearity
Eval= attacks his work, he believed in his theory so much even though he couldn' really back back it up with evidence
-
Experiment 1- view generalisation- clear evidence of non viewpoint generalisation, argues for structural description models. Effect of stereo depth information- 3D viewing condition- easier to perform in the extrapolated viewpoints, argues for image based models
Canonical perspective- experiment Palmer, Rosch and Chase- participants shown views of an object and are asked to rate how much each one looked like the objects they depict
in recognition task, reaction time correlate with the ratings, canonical views are recognised faster at the entry level
why- frequency hypothesis- its the view we mostly see in our life time. maximal information hypothesis- the viewpoint contains the most amount and most informative information about the object
Frequency hypothesis: the easiness of recognition is related to the number of times we have seen the objects from each viewpoint. Maximal information hpothesis= best views tend to show multiple sides of the object with all its parts- both are under canonical viewpoint
Faces:- organise our perceptual input, goes on without our conscious efforts. involves within category discrimination, errors in face recognition can have consequences e.g. eyewitness testimony e.g. Devlin 1976
Faces: Johnson and morton 1991- new born babies will preferentially view faces from day 1 (9 mins), expression analysis seems to be innate (Meltzoff nd Moore, 1977) though we already accept that this is independent of recognition
features vs Configural- feature- suggests faces are primarily remembered due to their facial features- Garner 1978, configurational (spacing)- places the emphasis on the relationship amongst the facial features- Bartlett and seared 1993 and diamond and Carey 1986
holistic hypothesis- whole, where both configurational and features info is required for accurate recognition- tanaka and sengco 1997
however, it does emphasis that loss of either rtpe of information could be detrimental to efficient face perception- carbon and elder 2005
upright faces- Yin 1969- whilst people are better at recognising upright faces than they are other objects, they are worse for invented faces than they are for other inverted objects
Thatcher illusion- Thompson 1980- we perceive faces in terms of the global configuration of facial deputes, we are unable to detect or process the properties of local individual face parts if upside down- evidence for holistically
model of face recognition- structurally encoded, activates face recognition unites, if match between encoding and FRU then semantic information can be accessed, personal identity nodes- contain info about that person- Bruce and young 1986
Using functional magnetic resonance imaging - subject view faces for a while, then picture s of objects, one area becomes more active during face viewing FFA, another area becomes more active during object viewing- LOC
-
facc recognition is special- infants show tendency to track moving faces, face agnosia (prosopagnosia) without object agnosia- object agnosia without prosopagnosia
-
inversion effect- healthy participants are better at recognising upright rather than inverted faces- this effect is not as strong with objects
-
Why do we see some things as objects? Edge perception, distance, tonal prosperities, texture
Gestalt psychology: Wetherimer, Koffa, Kohler- Circa 1910= Gestalt= German word that means 'whole'. The whole is other than the sum of its parts
Perceptual organisation- image or the visual system that allows you to correctly group different parts of the scene into objects? similarity e.g. colour, position, familiarity, Gestalt laws of perceptual organisation
Gestalt laws: Law of pregnazt e.g.good figure, law of simplicity- not really laws, just suggestions about what the visual system might be doing. They provide us some framework on how to separate figure from the ground e.g. figure ground separation
Jennifer anniston cell?- single cell recording in lie patients - very rate, patients were shown hundreds of photos- Quiroga et al. 2005 - neutrons in MTL medial temporal lobe responded only to Jennifer Aniston but not when she was photographed with brad phit, appears to be the Jennifer Aniston ell exists
but an important caveat, however, is that this pat of the brain is known to be involved in memory
meet the grebes? Gautheir and Tarr 1997- it is possible to train participants to become a grebe expert in a matter of hours, after this special training as they become expert- brain activity shifts from object area to face area. Specifically this experiment required subjects tolerant to identify 'grebes' which are weird shapes that are complex like faces but don't look much like them
face processing isn't special e.g. buyer et al 1983- prosopagnosic farmer could identify his cows, assail et al 1984- could recognise faces not now cows , McNeill and warrington 1933- patient with prosopagnosia who could distinguish between his sheep. ellis and young 1993- reflect specialities in processing for man types of objects
Summary: Gestalt principles- some insights on how figure/ground segmentation can be down. two main theories of object recognition e.g. image based models and structural description models. we have preferred stored viewpoint for an object- canonical view and grebes highlight that face recognition might only reflect expertise