Analysing images and human contrast sensitivity

Question 1

What is the task faced by the visual system?

Answer 1

To extract useful information about the environment from the patterns of light within the retinal image.

Question 2

How does Marr (1982) describe the task of vision?

Answer 2

Vision is an information processing task… The process of discovering from image what is present in the world, and where it is.

Question 3

What types of information do visual images contain?

Answer 3

Reflected light from objects which differ in terms of light wavelength and luminance.

Question 4

What do spatial variations in luminance in retinal images carry information about?

Answer 4

Object boundaries, edges and contours in the world.

Question 5

What are the different spatial scales at which luminance changes can occur?

Answer 5

Smooth (slow) -> medium -> abrupt (sharp)

Question 6

What do slow luminance changes reveal?

Answer 6

The coarse spatial structure of the world, e.g. Large objects and overall shape.

Question 7

What do abrupt luminance changes reveal?

Answer 7

The fine spatial structure of the world, e.g. Small objects and fine detail.

Question 8

How can visual information be represented in images?

Answer 8

• Through measuring the luminance at each spatial location in the image, like a digital camera, with points (pixels) determining quality. However this is inefficient.
• Through breaking the image down into basic components (building blocks, e.g. lines, blobs and corners). Sinusoidal gratings are a good choice - they capture the luminance variations in the image at each spatial scale.

Question 9

What are sinusoidal gratings?

Answer 9

Simple 1-dimensional, periodic patterns in which luminance varies across space.

Question 10

Describe how luminance varies across space in a sinusoidal grating.

Answer 10

Varies across the pattern (x) according to a sinusoidal waveform, across y it is constant.

Question 11

What are the defining features of sinusoidal gratings?

Answer 11

1. Spatial frequency (SF)
2. Contrast (intensity difference)
3. Orientation (axis of bars)
4. Spatial phase (relative position of the bars)

Question 12

Define spatial frequency (SF).

Answer 12

Spatial scale of luminance variation - number of cycles in one degree of visual angle.

Question 13

Why are grating patterns so useful?

Answer 13

It’s been shown mathematically that any image can be created from a set of sinusoidal grading patterns.

Question 14

When making a complex image from sinusoidal gratings, what do low and high SF gratings contribute, respectively?

Answer 14

Large objects/overall shape, fine detail/smaller objects.

Question 15

What is the process of creating images from sinusoidal gratings called?

Answer 15

Fourier synthesis.

Question 16

What is Fourier analysis, and how does it differ from Fourier synthesis?

Answer 16

Analysis is decomposing or representing an image as a set of sinusoidal gratings, synthesis is creating an image from gratings.

Question 17

What did DeValois and DeValois, 1990, show?

Answer 17

That Einstein’s face can be created using sinusoidal gratings of progressively higher SFs - 64=basic face shape, 164=clear, recognisable image.

Question 18

Why is coding with sinusoidal gratings economical?

Answer 18

Because unlike recording light intensity at thousands of different points, all you need to know is the 4 characteristics of each of the gratings.

Question 19

What relevance do sinusoidal gratings have to vision?

Answer 19

• they’re a ‘universal language’ to precisely describe visual scenes.
• we can measure the brain’s reaction to grating patterns (e.g. Contrast detection thresholds, cell firing rate)
• this can be used to predict responses to other images.

Question 20

What is MTF?

Answer 20

Modulation transfer function - the MTF of a system or cell is the extent to which each grating is transmitted.

Question 21

What is the assumption behind using MTFs to predict the visual system’s response to images?

Answer 21

All images are a sum of sinusoidal gratings, and the response to a complex pattern is basically the sum of its response to component grating patterns - assumption of linearity. If true, very powerful.

Question 22

Is the assumption of linearity in the visual system justifiable?

Answer 22

Maybe.
+ under many conditions the approach is successful.
- the brain violates certain mathematical assumptions.
Overall a decent estimate.

Question 23

How do you measure the MTF of the whole system for sinusoidal gratings with different SFs?

Answer 23

If our visual system does a good job of transferring certain SFs, we should need very little contrast to see those gratings, and vice versa. So we can measure psychophysically the contrast detection threshold for different SFs.

Question 24

What is the resulting MTF from the entire visual system’s sensitivity to gratings with different SFs known as?

Answer 24

The Contrast Sensitivity Function (CSF), plotted as a graph.

Question 25

Are we equally sensitive to all spatial frequencies?

Answer 25

No, we are relatively insensitive to very low and very high spatial frequencies.

Question 26

Give some examples of research investigating the CSF for human vision.

Answer 26

Campbell and Green, 1965; Campbell and Robson, 1968.

Question 27

What are the axes on a CSF graph?

Answer 27

y = contrast sensitivity (1/threshold contrast required to detect)
x = SF of grating patterns.

Question 28

For what SF of sinusoidal gratings is sensitivity greatest?

Answer 28

Gratings with spatial frequencies between 2-6 c/deg.

Question 29

Where are the patterns visible and invisible on a CSF graph?

Answer 29

Visible below the curve, invisible above it.

Question 30

What shape is a CSF graph?

Answer 30

Inverted U.

Question 31

What can we predict from the shape of the CSF?

Answer 31

Performance for more complex images, as they’re just a combination of grating patterns - if luminance variations occur at an optimal SF, they will be visible even when contrast is low. Also for very low and very high SFs contrast must be high to allow visibility.

Question 32

Why does the CSF have an inverted U shape?

Answer 32

• the fall-off in sensitivity at high SFs is due to optical imperfections in the eye which blur/degrade fine detail.
• the drop on sensitivity at low SFs is more difficult to explain, but is due to neural factors in the visual system’s information processing, rather than the eye itself.

Question 33

Exactly why is there a drop in sensitivity for low SFs?

Answer 33

Cells very early in the visual pathways, such as retinal ganglion cells, have receptive fields that exhibit a concentric centre-surround organisation (Kuffler, 2003). So when a similar luminance level covers the entire receptive field, they show little or no change in firing rate - they respond to gratings, but not when the SF is too low.

Question 34

What are the different luminance level ranges in which vision can be investigated?

Answer 34

1. Photopic - daytime
2. Mesopic - dusk
3. Scoptic - near dark

Question 35

What does the CSF look like at different light levels, according to Patel (1966) and De Valois et al. (1974)?

Answer 35

As the overall luminance level is progressively reduced, the peak sensitivity shifts to gratings of lower SF, occurring mainly because sensitivity to high SFs worsens.

Question 36

How do findings on CSF at different light levels relate to our experience?

Answer 36

It agrees - at night our rod system is most active and our ability to see fine detail is lost. Retinal ganglion cells are indirectly fed by rods and sacrifice acuity to operate at low light levels.

Question 37

What is a sinusoidal grating’s TF?

Answer 37

Temporal frequency - indicates how rapidly they move or flicker.

Question 38

What is temporal frequency measured in?

Answer 38

c/sec or Hz.

Question 39

What did Van Nes et al. (1967), Robson (1996), and Kelly (1979) investigate, and what did they find?

Answer 39

CSFs for moving/flickering gratings - sensitivity to very low spatial frequencies improves when the temporal frequency is high (e.g. 10Hz).

Question 40

What are the possible neural mechanisms behind increased sensitivity to low SFs when TF is high?

Answer 40

Magnocellular cells (primate lateral geniculate nucleus, thalamus) are 10x more sensitive to low SFs than parvocellular cells when patterns move or flicker at high rates (Derrington and Lennie, 1984; Kaplan and Shapley, 1982).

Question 41

What shape does the CSF have when the image is flickering?

Answer 41

Straight across, then down.

Question 42

What are the behavioural effects of lesions to the parvocellular layer of the LGN in monkeys?

Answer 42

Dramatic loss of sensitivity to stationary gratings, decreasing sensitivity as SF increases. Sensitivity at low SFs returns as TF increases.

Question 43

What are the behavioural effects of lesions to the magnocellular layer of the LGN in monkeys?

Answer 43

No effect on stationary gratings, but decrease in sensitivity for higher TFs at low SFs.

Question 44

Who studied the behavioural effects of lesions to cells in the LGN in monkeys?

Answer 44

Merigan et al., 1991.

Question 45

What do standard visual acuity tests try to work out?

Answer 45

The finest spatial detail that an observer can discern. Uses very high contrast stimuli of decreasing size.

Question 46

What are some examples of standard visual acuity tests, and what do they measure?

Answer 46

• Snellen eye chart - recognition acuity
• Landholt rings - resolution acuity
• Parallel bars - resolution acuity

Question 47

How are the CSF and standard acuity measures related?

Answer 47

• acuity tests measure the smallest details that can be resolved, with fixed and maximal contrast. Size limits performance.
• CSF tests measure how both size (SF) and contrast limit performance.

Question 48

Are CSF and standard acuity measures ever equivalent?

Answer 48

Yes - the highest detectable SF requires maximum contrast because sensitivity is so low, therefore this single point on the CSF is also a measure of visual acuity.

Question 49

What are the advantages of using visual acuity tests?

Answer 49

+ they’re mostly quick to administer - rapid assessment. Measuring CSF is time consuming and requires specialised equipment to create the gratings.
+ many causes of loss of sensitivity to fine detail are optical in nature - useful in a clinical setting, as can be alleviated by glasses or contacts.
+ deficits in sensitivity to lower SFs are neurological in origin, so cannot be alleviated by clinicians - there’s no need to measure the whole CSF.

Question 50

What are the advantages of measuring the CSF?

Answer 50

+ characterises sensitivity over all SFs, enabling scientists to predict the visibility of objects in any complex scene.
+ Glinsburg et al. (1982) used the CSF to predict how well pilots see objects - under conditions when fine detail is lost, e.g. fog, visual acuity is a poor indicator of performance.
+ some Alzheimer’s (Nissen et al., 1985) or cataracts (Hess & Woo, 1978) patients exhibit substantial sensitivity deficits for both low and high frequency patterns, which acuity tests would not pick up.
+ you get a measure of visual acuity anyway!