Recognition & Localisation Flashcards
Describe the hierarchical model of object recognition
- Detection of edges
- Detection of combinations of edges and contours
- Detection of parts of objects, such as a face
- Detection of object from a point of view
- Scale and orientation invariant detection of objects
- Categorisation of objects
What are the challenges to the hierarchical model of object recognition?
It doesn’t explain scale and orientation invariance fully. It doesn’t account for top-down feedback, eg, what is the cortex instructing the LGN? It is more likely that a population of neurons encodes a face than a single neuron as suggested by the hierarchical model.
Tell me about the arrangement of the Lateral Geniculate Nucleus
It is arranged into 6 layers, alternating in whether the input is from the contralateral or ipsilateral eye. The first 4 layers receive input from the parvocellular ganglion cells and last 2 layers receive input from the magnocellular ganglion cells. Their outputs converge onto either the parvocellular pathway or the magnocellular pathway. LGN projection neurons form 1:1 connections with the P/M ganglion cells, essentially acting as a thalamic relay station (LGN is in the thalamic nuclei). However, 60% of the LGN’s input is from the cortex, it is largely unknown what the function of this feedback is.
Tell me about columnar organisation in the ventral pathway
It is organised into layers and columns, the layers being different stages of visual processing in the hierarchical model. The columns are:
a) Ocular dominance columns, these correspond to inputs from either the contra or ipsilateral eye.
b) Orientation columns, they are key in detecting specific orientations.
c) Blob columns, key in the processing of colour.
Describe some methods to discover the columnar organisation
Ocular dominance columns were discovered and researched using radioactive proline injected into just one eye. Radioactive glucose can also be used, stimulating just one eye indicates the separation of contra and ipsilateral inputs.
Blobs were discovered using staining via cytochrome oxidase.
What is a simple cell?
A simple cell has a receptive field about 2-3x large than a LGN projection neuron, it responds to specific orientations if they’re on the centre of the receptive field. They are located in layers 4 & 6.
What is a complex cell?
A complex cell receptive field is made up of many simple cell receptive fields, given that they detect similar orientations. They can detect a specific orientation anywhere in the receptive field and are typically found in layers 2, 3 and 5.
What is a hypercomplex end-stopped cell?
They can recognise orientations anywhere within their receptive fields but will produce an inhibitory signal if it extends out of their receptive fields. However, if it changes orientation after it leaves the receptive field, there isn’t an inhibitory signal.
What is a grandmother cell?
AKA Jennifer Anniston neuron, can recognise faces and is scale and orientation invariant.
What is the orienting reflex?
It is the orientation of the head towards stimuli, salient stimuli are focused on the fovea, this response develops as we age, with none at 3 months but a strong orienting reflex at 10 months old.
What is smooth pursuit?
The movement of eyes to keep the moving object on the retina creating a smooth image. Aims to reduce the retinal slip, difference between where the targets motion and the eyes motion.
What are saccadic movements?
They are used in object inspection, the eye rapidly moves, anywhere from 0.5 - 30 degrees every 30-60ms focussing on a specific part of an object, for example, parts of a face such as the edges, nose, eyes, ears. It’s worth nothing that saccades are often followed by fixations where more detail can be taken up.
Why is object localisation important in hunting/prey capture?
Important in anticipating movement.
Tell me about the superior colliculus
The SC receives inputs from ganglion cells, the auditory system and somatosensory system. It is key in regulating saccadic movements. Cats with ablated SCs can’t orientate their heads towards stimuli.
It has a retinotopic map (adjacent spatial locations are represented by neighbouring neurons) made up of layers, each represented by different inputs from different brain areas, such as the retina, visual cortex and somatosensory cortex.
The layers can be visualized through soma staining.
What are the differences between mammalian and fish orientation pathways
Mammals use the Dorsal “where” stream:
M-Ganglion cells to LGN M pathway to V1 to V2 to V3 to Middle temporal to parietal lobe
Fish use the tectum, pretectum and hindbrain for their orientation responses.