M2, T1, object recognition processes and models Flashcards
Why is object recognition important?
- Recognise objects regardless of: Orientation, Distance, Lighting conditions
- Objects encountered everyday under a variety of circumstances
- Impaired viewing conditions: object recognition more difficult but can do it
- Two-dimensional images: photos, paintings i.e.. depth is absent
- Brain’s object recognition system automatically “copes” with a number of object transformations
Object recognition, description and storage
Object descriptions of structures of all known objects must be stored in the brain/mind
Object recognition involves a comparison of the seen object with the stored structural description of the object.
- what the item is, what it is used for, other objects it is associated with
Object recognition system - stored information
- Visual knowledge of an object’s shape
- Can process an object from various alternative viewpoints that it might be presented in
- Knowledge of semantic functions and object’s associates
- Knowledge of verbal names (labels) for objects
Models of object recognition
Marr’s (1980) model of object recognition
Ellis and Young (1988) model of object recognition
Riddoch and Humphreys (2001) model of object recognition
Marr’s (1980) model of object recognition
*look up image
Marr’s object recog model - Primal Sketch
- Initial object representation, initial stage
- Brightness changes across the field of vision allowing for perception of 2D geometry of images
- See edges because they result in sharp changes in brightness (luminance)
Marr’s object recog model - 2.5D Sketch
- Combines information such as textures and shading (depth information) from the primal sketch to construct and determine the spatial locations of visible surfaces from the viewers point of view
- Describes the object from only one view-point
Marr’s object recog model - 3D model
- Real shape of objects and surfaces and how they relate to one another
- Independent of view-point of viewer
- Could use this “standard” information to look up our list of known descriptions of objects for recognition
Ellis and Young (1988) model of object recognition
*look up image
Riddoch and Humphreys (2001) model of object recognition
*look up image
A hierarchical model of object recognition and naming, specifying different component processes which, when impaired, can produce varieties of apperceptive and associative agnosia
Ellis and Young object recog model - initial raw visual representation
Initial representation (Marr - primal sketch)
- intensity (brightness) changes across field of vision
- features such as edges, colours, etc
Ellis and Young object recog model - viewer centred description
Viewer centred representation (Marr 2.5D sketch)
- spatial locations visible surfaces
- uses information such as depth and location
- object only represented from the viewpoint of viewer
Ellis and Young object recog model - object centred description
Object centred representation (Marr - 3D model)
- real shape of objects and surfaces
- independent of view-point of viewer
- could use this ‘standard’ information to look up object in object store
Ellis and young object recog model - object recognition units
Object recognition units
- stored structural object representations
- comparison of viewer and object centred representation
- interface between visual and semantic object representations
- one recognition unit for each known visual object
- describe what an object looks like
Ellis and young object recog model - semantic system
Semantic representation
- Describes object properties and attributes
- Object Recognition Units access semantic representations if match between visual representation and stored object description at levels of object recognition unit
- Access to Semantic representation occurs at number of levels
- Number of different types of input gain semantic access
object, picture, written name, spoken name
Ellis and Young object recog model - speech output lexicon
- Name codes for objects not contained within semantic system
- Object Name is represented in separate store – Lexicon
- Must access object name via semantic system
- Object naming task is typically used with neuropsychological patients
Ellis and Young object recog model - Perceptual Classification Stage Evidence
- Perceptual classification involves matching viewer-centred and object-centred representations
- Repetition Priming: items, which have been previously encountered at an earlier phase of an experiment, are more easily recognised than those that have not been previously encountered
- Repetition Priming effects are modality specific
Warren and Morton (1982) - priming
Briefly presented pictures of objects – task naming aloud
- Found that subjects were able to identify a picture of an object at a shorter exposure duration if they had either …
- > Seen the same object in a picture previously or
- > A different picture of the same object previously in the experiment
- prior exposure to object label had no effect on recognition performance
- repetition priming is modality specific - only occurs if first and second presentation appear in same modality
Ellis and Young object recog model - semantic processing stage
Semantic System: stage at which object’s functions and associates are retrieved from memory
Access to semantics occurs – object seen, heard, felt or if written name or spoken name encountered
Why would this be important?
Semantic Priming – effect of related item context on performance i.e., presenting related items facilitates responding to the second item
-> e.g. nurse - butter slower than bread - butter
Ellis and Young object recog model - Name retrieval stage/speech output lexicon
- Access names of objects (verbal labels) via semantic system
- Objects and pictures have direct access to their meanings but an indirect access to their labels which occurs via semantics
- > classification of pictures faster than decisions requiring access to object labels (Potter & Faulconer, 1975)
- > word recognition – words named faster than they are categorised
- > interference effects – irrelevant names interfere with object naming but not with picture classification (Glasser & Glasser, 1995)