Digital Images & Point Processes

Question 1

How do you acquire an image?

Answer 1

Using the source of energy (light normally), you use a sensor to gather the reflected light, and thereby form an image.

Question 2

What is sampling and what does it determine?

Answer 2

Digitisation of the spatial coordinates
Determines spatial resolution

Question 3

What is quantisation and what does it determine?

Answer 3

Digitisation of the light intensity function
Determines grey level, colour or radiometric resolution - in rough terms, how much of the light energy is quantised

Question 4

What is the Nyquist rate?

Answer 4

Samples must be taken at a rate that is twice the frequency of the highest frequency component to be reconstructed

Question 5

What is under-sampling, and what happens when it occurs?

Answer 5

Under-sampling - sampling at a rate that is too coarse i.e. below the Nyquist rate
When it occurs, aliasing becomes prominent in the image, which are artefacts that result from under-sampling

Question 6

How does aliasing occur?

Answer 6

Aliasing occurs when two signals become indistinguishable when sampled
Can also be introduced when the image is resampled

Question 7

What is quantisation?

Answer 7

Determines the number of levels of colour/intensity to be represented at each pixel

Question 8

What is anti-aliasing?

Answer 8

Smoothing out of high-frequency signals before sampling so its impossible to ‘see’ the alias

Question 9

How do you downsample an image, and when does it happen?

Answer 9

When downsampling, you need to compute a summary pixel value from each local area i.e. take the mean value (as a whole number) from the surrounding elements e.g. 25, 26, 26, 27 gets combined into one square, which is 26

This happens when re-sizing an image (making it smaller only)

Question 10

How do you upsample, and when does it happen?

Answer 10

When upsampling, you need to interpolate from the known values to produce an estimate at the unknown pixels
Average the known values in a local region centred on each unknown pixel, fit some kind of function through known values

Occurs when re-sizing an image to make it bigger.

Question 11

What is re-quantisation?

Answer 11

Grey level resolution (quantisation) can be dropped by dividing each pixel value by a constant. There is a side effect - can’t increase grey level resolution of a single pixel

Question 12

How do you acquire colour images?

Answer 12

Using the RGB model, and a single CCD which uses the Bayer pattern.

Question 13

What is the Bayer Pattern?

Answer 13

A grid of each different colour from the RGB model.
One colour value is measured, two are estimated at each pixel
Typically, one colour is measured using their neighbours’ mean, whereas the other colour is measured using their diagonals’ mean

Question 14

What are some benefits of working with colours in greyscale over RGB?

Answer 14

Makes processing easier
Reduces the amount of information
Makes some of theory simpler
Many image processing methods were developed for single value images, of which the value was usually intensity or grey level

Question 15

What is the formula for converting between RGB images and greyscale images?

Answer 15

i = 0.3R + 0.59G + 0.11B

Question 16

What is the formula for converting to a Green-weighted image?

Answer 16

i = G - (R+B)/2

Question 17

What is HSV, and what is it based on?

Answer 17

HSV is an acronym, which stands for its three parts:
Hue - What general colour it is, indicates the angle around the colour wheel
Value - How bright or dark it is, indicates how dark light the colour is
Saturation - How strongly coloured it is, indicates how strong the colour is

Question 18

What is an intensity transform?

Answer 18

Take the source pixel, transform it in some way, then overlay it onto the target pixel

Question 19

What is a linear transform?

Answer 19

Two commonly used point processes are multiplication by and addition of a constant i.e. g(x,y) = a * f(x, y) + b
A is the gain and controls contrast, B is the bias and controls brightness

Question 20

What is negation?

Answer 20

Basically, just flipping everything on its head
Inverting all the colours

Question 21

What is dynamic range?

Answer 21

A set range for the pixels to exist within.
Intensity transforms require data sometimes to be within a certain range.

Question 22

What is contrast stretching?

Answer 22

Converts a source image in which intensities range from min s to max s, to one in which they range from min t to max t

Question 23

What is an example of non-linear transformation?

Answer 23

A common example is thresholding

Question 24

What is grey level slicing?

Answer 24

This highlights a specific range of intensities, whereby you can then reduce other grey levels to a lower level, or preserve them.

Question 25

What is gain?

Answer 25

Gain is the level of contrast

Question 26

What is the bias?

Answer 26

Controls the level of brightness