Revision Qs

Question 1

To make fine details more visible you would..

Answer 1

Use negation

Question 2

To pick out a particular intensity you would…

Answer 2

use grey level slicing

Question 3

To reduce salt and pepper noise you would….

Answer 3

Use median filtering

Question 4

Why is median filtering so good for salt and pepper noise?

Answer 4

target pixel gets median value so unlikely to replace noise with more noise or a good pixel with noise

Question 5

To reduce gaussian noise you would…

Answer 5

Use gaussian filtering

Question 6

To blur an image but preserve edge detail you would…

Answer 6

Use gaussian filtering

Question 7

To enhance edges in a smooth image you would…

Answer 7

Use an unsharp mask

Question 8

How would you detect strong edges with 2nd order derivatives?

Answer 8

Using a Laplacian filter

Question 9

To increase brightness you would…

Answer 9

Use bias (shift intensity)

Question 10

To increase contrast you would…

Answer 10

Use gain (contrast stretching)

Question 11

To reduce all pixel values above a certain level to white and the rest to black, you would..

Answer 11

Use thresholding

Question 12

To make text easily stand out against the background you would…

Answer 12

Use thresholding

Question 13

To correct an image that looks washed out due to non linear lighting effects you would…

Answer 13

Use gamma correction

Question 14

To sharpen an image slightly without introducing much noise you would…

Answer 14

Use an unsharp mask

Question 15

To reduce small random noise evenly across an image, you would….

Answer 15

Use a mean or gaussian filter (gaussian would preserve your edges more)

Question 16

From an intensity frequency graph, how would you calculate PDF?

Answer 16

freq / number of pixels - this is just probability

Question 17

How do you calculate CDF from PDF in a table?

Answer 17

Starting at the smallest intensity, you add a running sum of PDFs. The first CDF will equal its PDF and the last will be 1

Question 18

What is the histogram equalisation formula?

Answer 18

Equalised pixel output =
round( (CDF - CDF_min) x (L - 1)/(1 - CDF_min) )

Question 19

What does binariasation mean?

Answer 19

Thresholding an image so it only contains two colours, typically black and white

Question 20

What is Otsu thresholding?

Answer 20

A method that determines the threshold using the images bimodal histogram

Question 21

When is Otsu thresholding appropriate?

Answer 21

On an image with a bimodal histogram, e.g. a peak of light pixels and a peak of dark pixels

Question 22

How is the threshold found for Otsu?

Answer 22

By minimising a weighted sum of the variations from the two histogram regions to ensure the threshold is splitting them evenly

Question 23

When is unimodal thresholding appropraite?

Answer 23

For histograms with one peak, e.g. an image with a lot of dark pixels or a lot of light pixels

Question 24

How is the threshold found in unimodal thresholding?

Answer 24

A line is drawn on the histogram from the highest peak to the lowest bin, and then the bin is selected who’s peak is furthest from the line

Question 25

What is Unimodal best for?

Answer 25

Images with high intensity gradients

Question 26

When might you apply a thresholding method locally?

Answer 26

When the image is not uniform but sections of the image (strips, quarters) adhere to the conditions for a method.

Question 27

What does it mean to apply a method adaptively?

Answer 27

To apply a method based on the properties of the given image, using local regions determined for that image

Question 28

You have tiny thin lines connecting objects in an image, what could you try?

Answer 28

Opening, it will erode thin connections then dilate to restore shapes

Question 29

To remove tiny gaps in letters in text, you would...

Answer 29

Use closing to fill the gaps and solidify letters

Question 30

To remove lots of white dots scattered on an image background, what morphological methods could you use?

Answer 30

Use opening or closing to remove small isolated foreground pixels

Question 31

What types of Sobel operators are more stable in noise?

Answer 31

Larger ones

Question 32

Whats a downside of the roberts operator?

Answer 32

low resistance to noise

Question 33

What kind of edge size do roberts operators give?

Answer 33

Thin

Question 34

When would you choose the Roberts operator over Sobel or Canny?

Answer 34

For fast, simple edge detection with low cost and minimal noise concern

Question 35

Why is Sobel generally a preferred operator over Roberts?

Answer 35

Sobel has better noise resistance and directional edge information because of its larger neighbourhood

Question 36

In practice how are roberts and sobel operators applied?

Answer 36

By placing the operator over the appropriate amount of pixels, multiply it element wise for both x and y then computing a gradient. - In roberts replace the whole area, - In sobel, replace the target pixel

Question 37

In practice why is canny edge detection used rather than just sobel?

Answer 37

Canny includes multiple steps: noise reduction, gradient calculation, non-maximum suppression, and hysteresis thresholding, which together produce cleaner, thinner, and more connected edges.

Question 38

What pre-processing step is crucial before using any derivative-based edge detection method? What is a typical method?

Answer 38

Smoothing or denoising the image - e.g. with a gaussian blur

Question 39

How do you convert the results of Roberts or Sobel into a displayable image?

Answer 39

Normalise the gradient magnitudes to the range 0,255

Question 40

What is an advantage of split and merge over region growing?

Answer 40

Does not require a predefined seed points and can handle a global structure better

Question 41

What is the main limitation of using basic k-means for segmentation?

Answer 41

It doesn't consider spatial connectivity, so segmented regions may not be spatially coherent or aligned with actual object boundaries.

Question 42

Why might combining region- and edge-based methods be beneficial?

Answer 42

To leverage both intensity similarity (region) and boundary accuracy (edges), improving robustness.