Lecture 4 - Spacial vision: Lines and textures

Seite 1

Chapter 3

Area V1 (also called the striate cortex, the primary visual cortex, or area 17)

As its name implies, it has stripes. These stripes are caused by the multiple layers which make up area V1. These layers each receive information from different sources (such as cells P, M, and K of the LGN, information from both eyes still being kept separate) and also send information to different sources (areas of the cortex, feedback to LGN). Aside from the areas whch receive information from the LGN, which keep both eyes' information separate, we can also find binocular cells, which begin to put the data from both eyes together. They are about 70% of area V1 and, in following areas, most cells are binocular.

Orientation selectivity

Cells in area V1 are tuned for orientation. While cells in the retina and LGN react to any spots of light, cells in area V1 are very picky about the stimuli they respond to. They seem to prefer lines, but certain cells seem to respond to lines of certain orientations. What stimuli a cell responds to tells us a lot about its tuning.

When a cell is tuned to a stimulus, it responds most intensely to that specific stimulus, less strongly to stimuli similar to it, and not at all to others. The wider the cell's bandwidth, the more further away from it's preferred stimulus it can activate. These kinds of cells are called filters.

Organisation of the primary visual cortex

The retinotopic mapping we find in the LGN (the "image" formed on the retina is maintained in the same "shape" with the organisation of cells) can also be fount in V1. Each V1 (one in each hemisphere) maps half the visual field. However, the numbers of cells dedicated to the fovea is significantly higher than the cells dedicated to the peripheral vision.

Inside area V1, there are multiple cells dedicated to certain orientations. They are organised sensibly, however, so we end up with orientation collumns, each with a slightly different orientation than the one next to it. Eventually, all orientations are covered and you get a hypercolumn.

A hypercolumn, however, only analyses one area of the visual field, so we need multiple hypercolumns to be able to detect orientations throughout the entire field. There are less hypercolumns in our peripheral vision than the fovea area. Therefore, when we see a line in a particular area only the cell(s) sensitive to its orientation which cover that area will fire.

Therefore, when you're looking at an eye, the cells in V1 will only fire for the contours that match their orientation.

Seite 2

Simple cells

While cells in the LGN can distinguish edges, they are indifferent to orientation. This is where cells in V1 come into play.

Simple cells have both ON and OFF fields, but they are next to each other and stretched out, as opposed to the cells in the LGN and retina.

These cells can either be bar detectors and edge detectors, but otherwise function in the same way.

To make these orientated cells using data coming from the LGN, we might use, at a 45 degree angle for example, the following method: for the ON region, we take output from ON centre cells aligned in a 45 degree line, while for the off region(s) we take data from OFF centre cells in the corresponding 45 degree areas.

Considering that simple cells have both ON and OFF areas, it is easy to assume that they are sensitive to the position of the line/edge in their receptive fields (phase sensitivity).

Seite 3

Complex cells

Complex cells detect the same lines and edges as simple cells, except they are phase insensitive (they don't care about the positioning of the line in their receptive field). They can signal both excitatory and inhibitory signals at any point in their receptive field, depending where the line is, so it is harder to guess their tuning.

It is theorised that complex cells use data from simple cells to determine their output. A coomplex cell would be connected to a number of simple cells with the same orientation, covering a certain area. Wherever the line appears, within this small space, one of the simple cells should give a good response and therefore this is all that is needed by the ‘OR gate’. Hence, it gives a response to all positions within this receptive field.

Hypercomplex cells

The further away from the eye we go, the more specialised the cells become. Like complex cells, hypercomplex cells are tuned to a specific orientation. However, while complex cells react the same way to a bar that covers their receptive field and one that exceeds it, hypercomplex cells do not. They, in turn, become inhibited by lines longer than their receptive fields (like having an OFF zone at the edges). Therefore, hypercomplex cells are sensitive to both a specific orientation AND length of a bar/edge.

We could say that hypercomplex cells form their ON areas by gathering data from corresponding excitatory complex cells while other cells, at the edges, produce an inhibitory response.

Seite 4

Face cells

Trigger features (conditions that need to be met for the cell to fire) become more and more complex as cells advance through the visual cortex.

Many studies seem to suggest that we have cells dedicated to recognising faces. Even studies done on sheep seem to indicate that they have special "sheep face" cells. This might mean that there is something special about the way we perceive faces.

The grandmother cell hypothesis

Since it was discovered that we might have specific cells dedicated to faces, a hypothesis rose, the Grandmother Cell hypothesis, which claims that we have a cell tuned to everything we perceive, including our grandmother. However, this cannot possibly be the case, as we would need a cell for every single thing we've seen and we will ever see. Therefore, there must be a limit to this tuning. But we aren't sure where the limit is yet.

Area V1 does, however, have cells dedicated to orientation, direction, colour, binocular disparity and so on.

Seite 5

The extrastriate cortex

Past area V1, there are a total of 30 visual areas in the extrastriate cortex.

While all these areas might have some specialised role in vision, it is important to note that they all heavily interact, and specialisation isn't as simple as it sounds.

From V1, information is sent to V4 (ventral - the "what" pathway) and to V5 (dorsal - the "where" pathway).

When cells from area V5 receive information, we might expect them to do something similar to hypercompex cells. However, they seem to have a different role altogether. They usually detect movement, regardless of shape or orientation of the line.

V4 cells may, similarly, detect something different like colour.

However we do not know how our brains put this information all together :(.

Seite 6

Illusory contours

In some cases, such as the Kanizsa triangle (three "Pac-man" shapes facing towards a central point, their "mouths" seemingly forming the corners of a triangle when there is nothing really there), we might see edges when they do not actually exist. These type of fake edges are called illusory contours. Might be explained through the effect hyperactive cells have, creating the illusion of an overlapping triangle over the circles. Illusory contours are also experienced by other animals, such as baby chicks.

Seite 7

Chapter 4

The tilt after-effect

The tilt after-effect is good evidence that we have orientation-selective neurons. Neurons sensitive to the specific orientation that we are being shown adapt over time, and therefore when we move our gaze to an orientation that is similar, the cell tuned to it will still have maximum excitation, but the ones surrounding it (which should give about 50% excitation) have their ratio fucked up, because one of them is inhibited, therefore the line looks slightly tilted in the opposite direction.

Tilt threshold

As expected, low-contrast lines are harder to detect for the orientation sensitive neurons, and therefore they fire less. If a cell has adapted to a certain orientation, it becomes harder to see. Therefore, if we stare at vertical black and white lines for a while, a very faint picture of the same type of lines would become invisible.

The size after-effect

Aside from orientation-based cells, we also have cells with different sized receptive fields (and orientation based), to account for different sized stimuli. The effect created, after adaptation to a certain thickness/size, is similar to the tild after-effect and also respects a distance paradox. The lines thinner than the original stimulus look fatter and the thicker lines look skinnier.

Simultaneous tilt and size effects

We can also produce simultaneous versions of the tilt and size after-effect. These are very similar to the simultaneous contrast illusion, except with orientation and size sensitive cells. This type of comparison might also be responsible for the Titchner or Ebbinghaus illusion. In this illusion, two identical sized circles are placed one between bigger circles and the other between smaller ones. Of course, the first one looks smaller than the latter.

Size threshold

As with the tilt, after adapting to a certain size stimuli, our cells become inhibited and stop firing as much. Therefore, the threshold is raised and similar patterns are difficult to see.

TERMINOLOGY: Bars are characterised in the research world by how many of them would fit in a certain space (spatial frequency). Therefore, a pattern with a high spatial frequency has lots of thin lines and one with a low spatial frequency would have less lines, but thicker.

Seite 8

Contrast sensitivity

One way to test which type of bars we are most sensitive to is by decreasing the contrast between the bars and the background and seeing how similar we can make them before the bars disappear completely. Blurring the edges of the lines is necessary. If we do this to a picture of lines with gradually increasing spatial frequency, we will notice that lines along the edges of the image start to blur out as the contrast increases, while lines with medium spatial frequency tend to stick around for longer. This shows that we are more sensitive to medium sized likes than to big or thin ones.

Lines under the resolution limit (also called visual acuity) are too thin (and/or numerous) for us to even see (basically everything that is too small for the human eye). Also, lines above the graph curve of the spatial contrast sensitivity function are invisible to the eye, if presented with that level of contrast or lower. So all the information that lies above our resolution limit is invisible to us, but so is any information that lies at lower spatial frequencies if its contrast is too low. The spacial contrast sensitivity function describes our window of visibility. Anything outside of the curve is invisible to us.

In most images, there is information belonging to many spatial frequencies. Our brains need channels (groups of cells tuned to the same requirements) at lots of spatial freq's to get all of this information. Channels can be tuned to all characteristics you have learned so far. All the images we see are split up into channels which analyse information at a different scale, but we do not perceive this. All we see is a complete image.

Peripheral vision

We are more and more unable to see high spatial frequency patterns the further away we go from the fovea. However, low frequency patterns seem to be unaffected. This points toward the theory that our peripheral vision is the same as our foveal vision, only coarser and, by simply magnfying the stimuli in our peripheral vision, we can fix that. It is, indeed, exactly what happens. By making letters in our peripheral vision larger, we are able to focus on them and read them as easily as we would in our foveal vision.

Seite 9

Chapter 8

Contrast sensitivity in children

We know young children have blurry vision. Aside from their resolution limit (acuity) being way lower than ours (not able to see thin lines), we can also see that they cannot recognise medium lines if the contrast is low enough. This curve is way lower than the normal adult curve in the spatial contrast sensitivity function.

Aside from normal acuity, we also test babies in Vernier acuity, which tests how small of a displacement in a line they can detect. They start off absolutely shit at this, way worse than normal acuity, but it gets better rapidly. In adults, we can be up to 10 times better at Vernier acuity than the other one. So the two types of acuity develop at very different rates. This must be because different brain areas develop at different ages, and the two processes seem to not depend on the same area.

Orientation perception in babies

Children seem to have some knowledge of orientation from birth, but if the stimuli are flashing, that ability seems to disappear. Only at 6 weeks old do the infants seem to be able to distinguish between orientations even when the stimuli are not constant.