Please enable JavaScript.
Coggle requires JavaScript to display documents.
Mid-Level Vision - Coggle Diagram
Mid-Level Vision
.
- A large proportion of the brain is devoted to vision; over 20 different richly interconnected areas
- Dorsal pathway (where/how) > a major visual pathway that runs from V1 and V2 into MT (V5) and then to the parietal cortex; represents properties that relate to motion or location of objects that can be used to guide actions.
- Ventral pathway (what) > A major visual pathway that runs from V1 and V2 into V4 and then to the inferotermporal cortex; represent properties that relate to identity of objects, such as colour, shape.
Visual processing occurs in numerous parallel streams:
- different processing channels parse the incoming signals into parallel streams (specialised for different attributes) to provide a segregated but compact and efficient input to the brain; observed at every stage of visual processing (retinal, subcortical (LGN), primary visual cortex, dorsal and ventral stream)
Modular organisation: he visual brain consists of many different visual areas, which are functionally specialised to process and perceive different attributes of the visual scene.
- Ultimately, the parallel input signals must be integrated into the cortex to provide a unified and coherent percept (the binding problem).
hierarchical organisation
- Visual processing is hierarchically organised: sequential processing stages increase in processing complexity from early to later visual brain areas.
- The stimuli required to produce optimal activation of a neuron become more complex as the visual processing stages progress along the processing hierarchy.
- It is not always clear tuning to complex patterns emerges from tuning to simpler properties in earlier visual areas (orientation, spatial frequency, colour)
- Progressively larger receptive field size from earlier to later visual areas
- Interestingly, neurons in higher visual areas also have progressively larger receptive field sizes, presumably aiding the integration of simple, spatially separated features into more coherent and organised perceptual units.
Inferotemporal cortex: parallel organisation and functional specialisation
- Part of the “what” pathway, situated in the lower portion of the temporal lobe.
- Very large receptive field sizes (some cover half of the visual field).
- Columnar organisation: cells that respond to similar features cluster in columnar regions that are elongated vertically to the cortical surface.
- The selectivity of cells within a column is similar but not identical, as exemplified by similar but different icons within each column.
Three clearly different areas for cell response:
- FFA: Fusiform face area (responds to faces more than other objects)
- PPA: Parahippocampal place area (responds preferentially to places, such as pictures of houses)
- EBA: Extrastriate body area (specifically involved in the perception of body parts)
.
Neuropsychological evidence for the modularity in the ventral visual pathway: the effect of brain damage in IT cortex.
- Huge individual differences in agnosias.
Associative agnosia
- Prosopagnosia
- Disorder of face perception in which the ability to recognise familiar faces, including one’s own face (self-recognition), is impaired, while other aspects of visual processing (feature detection, object recognition, object discrimination) is unaffected.
- Object agnosia
- Impaired recognition despite spared perception of basic features such as shape, size, colour, and orientation.
Integrative agnosia
- A form of visual agnosia in which individuals are able to perceive the elements of an object but find it difficult to perceive spatial relations between different parts and to integrate disparate parts of visual objects into perception wholes.
- Individuals affected by this kind of agnosa can usually draw an object accurately but only through a laborious process.
Apperceptive agnosia:
- Impairment in both low-level perception and recognition of objects
- Typically, a widespread damage to the alteral occipital and/or parietal-occipital areas of the brain.
- An inability to copy and match shapes despite normal visual field, normal visual acuity, normal colour vision, normal depth perception, normal eye fixations, and intact movement perception.
- Simultanagnosia is the inability to perceive more than one object at a time. Patients are able to identify individual elements of a complex scene but have great difficulty in understanding what is occurring overall within the scene.
- Ishihara colour plates can be used to detect simultanagosia > not due to lack of colour vision but rather an inability to perceive the individual coloured dots in the Ishihara figures as a collective whole.
- Will be more issues if damage in right hemisphere.
Implications of neuronal selectivity and functional specialisation in IT cortex for theories concerning object recognition
- Grandmother cell> a highly specialised hypothetical neuron that responds to one specific visual stimulus and no other, such as the image of one’s grandmother.
- Problems:
- How would the grandmother cell cope with changes in viewing conditions
- Too many different stimuli to assign specific neurons
- Most neurons respond to a number of different stimuli.
- These cells are not located in the Inferior temporal lobe but in the medial temporal lobe > essential for memory and concepts.
- Population (distributed) coding: combination coding > a pattern of firing across many different neurons as a code for specific objects; a large number of stimuli can be coded by a few neurons.
.
Models of object recognition
- Templates: the visual system recognises objects by matching the neural representation of the image with a stored representation of the same “shape” in the brain.
- Structural descriptions: recognition by components: the visual system uses a description of an object in terms of the nature of its constituent parts (geons) and the relationships between those parts.
perceptual organisation
- Successful region segmentation (determining what belongs with what to form coherent objects and units) involves linking some elements and excluding others to form a unit.
- Segmentation processes apply to both scenes (figure-ground) and objects (part segmention)
- Part segmentation (parsing) dividing a single object (element) into parts
- Perception of objects gives rise to the spontaneous perception of parts that are stable and nonarbitrary.
- Importance of points of high concavity: the visual system divides objects into parts where they have abrupt changes in surface orientation toward the interior of the object.
- Mid-level is synonymous with the notion of perceptual organisation
- Gestalt school: established in the first half of the 20th century;
- Gestalt psychologists have foreshadowed the area of research we now refer to as mid-level vision
- They have also proposed all known (and working) solutions to these problems.
- “The perceptual whole is different than the sum of its parts”
- Vocal critique of structuralism (a number of sensations add up to create our perceptions)
- Instead it was proposed that perception is not built up from sensations, but is a result of perceptual organisation.
- We see perceptual groups before we see individual elements; perceptual processes proceed from the global to the local > global superiority effect.
- Local is faster and interferes with the recognition of the global
- Gist perception: global properties that describe the spatial layout and the general content of the scene can be extracted extremely quickly (under 50 ms) well before its individual elements are processed and scrutinised.
- Global image features of scenes that are considered holistic and can be perceived as extremely rapidly include
- Degree of naturalness, openness, expansion, and colour
- Gist perception is thought to be based on rapid computations of the average spatial properties over the entire visual field > average orientation, average colour, average spatial frequency, etc, presumably with neurons with large receptive field sizes (though we do not know at what levels of a visual hierarchy these neurons might bel perhaps at different levels, depending on the gist dimension).
- Figure-ground processing in V1
- The receptive fields (RFs) of neurons in primary visual cortex (V1) are specialised to conveying information of a low-level nature, such as the orientation and spatial frequency of luminance contrast. They also respond to a very limited portion of the visual field.
- V1 neurons can be affected by information about the perceptual interpretation of larger parts of the scene than covered by the RF.