convoloution and image filtering

Question 1

What is image filtering?

Answer 1

It is the process of modifying or enhancing an image by applying mathematical operations to its pixel values.

Question 2

What are the two main types of image filters?

Answer 2

Point-based filters and neighbourhood (spatial) filters.

Question 3

What is a point-based image filter?

Answer 3

A filter that processes each pixel independently, without considering neighbouring pixels.

Question 4

What is a neighbourhood filter?

Answer 4

A filter that uses a surrounding region (kernel) of pixels to modify each pixel based on local context.

Question 5

What are common goals of image filtering?

Answer 5

Denoising, deblurring, contrast enhancement, edge detection, and feature extraction.

Question 7

What does brightness adjustment do in point filtering?

Answer 7

It adds a constant to all pixel values, making the image lighter or darker.

Question 8

What is a visual effect of brightness adjustment on a histogram?

Answer 8

It shifts the histogram left (darker) or right (brighter).

Question 9

What does contrast enhancement do in point filtering?

Answer 9

It stretches pixel intensities away from the center to increase tonal separation.

Question 10

What is the formula for linear contrast stretching?

Answer 10

I’(x, y) = a · (I(x, y) - 128) + 128

Question 11

What is image inversion in point-based filtering?

Answer 11

It replaces each pixel value with 255 minus the original, flipping dark and light.

Question 12

What does image inversion do to the histogram?

Answer 12

It flips it horizontally around the midpoint.

Question 14

What is quantisation in imaging?

Answer 14

It is the mapping of continuous pixel values to discrete levels due to limited bit depth.

Question 15

What is the bit depth in an 8-bit image?

Answer 15

Each channel has 256 intensity levels (0 to 255).

Question 16

Why is quantisation important in image filtering?

Answer 16

Because it determines the precision and dynamic range available for processing.

Question 17

What is the limitation of point-based filters?

Answer 17

They cannot recognize local patterns or remove structured noise like salt-and-pepper.

Question 19

What is an image histogram?

Answer 19

A graph that shows the distribution of intensity values across all pixels in an image.

Question 20

What do the X and Y axes in an image histogram represent?

Answer 20

X = intensity levels; Y = number of pixels with that intensity.

Question 21

What does a narrow histogram indicate?

Answer 21

Low contrast — most pixels are near a similar brightness level.

Question 22

What happens to the histogram after contrast stretching?

Answer 22

It expands to cover more of the intensity range.

Question 24

What is a box filter?

Answer 24

A neighbourhood filter that averages all pixels in a local window with equal weight.

Question 25

What is a Gaussian filter?

Answer 25

A neighbourhood filter that gives more weight to the central pixels and less to distant ones using a Gaussian distribution.

Question 26

Why is a Gaussian filter better than a box filter for blurring?

Answer 26

It preserves edges better by reducing the influence of distant pixels.

Question 27

What is the purpose of using a kernel in neighbourhood filtering?

Answer 27

The kernel defines how each neighbour contributes to the output pixel.

Question 29

What is convolution in image processing?

Answer 29

A mathematical operation that combines a kernel with an image to compute a new pixel value.

Question 30

What is the formula for 1D discrete convolution?

Answer 30

(f * h)(x) = Σ f(k) · h(x - k)

Question 31

What is the formula for 2D discrete convolution?

Answer 31

(f * h)(x, y) = Σ Σ f(m, n) · h(x - m, y - n)

Question 32

What does convolution do in practice?

Answer 32

It applies a weighted sum of neighbouring pixels to compute each output pixel.

Question 34

What are the key properties of convolution?

Answer 34

Commutative, associative, and distributive.

Question 35

What does the commutative property of convolution mean?

Answer 35

The order of image and kernel doesn’t matter: f * h = h * f

Question 36

What does the associative property of convolution mean?

Answer 36

Multiple convolutions can be grouped: (f * g) * h = f * (g * h)

Question 37

What does the distributive property of convolution mean?

Answer 37

f * (g + h) = f * g + f * h

Question 39

What does a Gaussian kernel look like?

Answer 39

A symmetric matrix where values are highest at the center and decrease smoothly outward.

Question 40

What is the purpose of a Gaussian low-pass filter?

Answer 40

To blur or smooth an image while preserving important edges.

Question 41

Why is convolution better than point filtering for noise removal?

Answer 41

Because it uses neighbourhood context to average or suppress random fluctuations.

Question 42

What visual effect does convolution with a small kernel produce?

Answer 42

Slight blurring or smoothing of the image.

Question 43

What happens when the convolution kernel contains negative values?

Answer 43

It can perform edge detection or sharpening depending on the kernel design.

Question 45

What is the difference between filtering and convolution?

Answer 45

Filtering is the general concept; convolution is one method of implementing filtering.

Question 46

What is the effect of using a larger convolution kernel?

Answer 46

More blurring or averaging, with increased computational cost.

Question 47

What are real-world uses of convolution in visual computing?

Answer 47

Noise reduction, image enhancement, edge detection, and neural networks.

Question 48

What is stride in convolution?

Answer 48

Stride determines how many pixels the kernel shifts at each step during convolution.

Question 49

What does increasing the stride do?

Answer 49

It reduces the spatial resolution of the output, effectively downsampling the image.

Question 50

What is the output size formula when using stride?

Answer 50

Output size = floor((Input size - Kernel size) / Stride) + 1

Question 51

If the input is 5x5 and the kernel is 3x3, what is the output size for stride = 1?

Answer 51

3x3

Question 52

If the input is 5x5 and the kernel is 3x3, what is the output size for stride = 2?

Answer 52

2x2

Question 54

What is padding in convolution?

Answer 54

Padding adds extra rows and columns around the input to control output size and preserve borders.

Question 55

Why is padding used?

Answer 55

To prevent shrinking of image dimensions and preserve spatial resolution.

Question 56

What is a common padding size for a 3x3 kernel?

Answer 56

1 pixel on each side

Question 57

What happens if no padding is used?

Answer 57

The output image becomes smaller than the input.

Question 58

What is the formula to maintain output size equal to input size in convolution?

Answer 58

Padding = floor((Kernel size - 1) / 2)

Question 60

What is dilated (atrous) convolution?

Answer 60

A convolution that introduces gaps between kernel elements to expand the receptive field.

Question 61

Why use dilated convolution?

Answer 61

To cover a larger area without increasing the kernel size or number of parameters.

Question 62

How does dilation affect the kernel?

Answer 62

It spreads the kernel elements apart while maintaining the same number of weights.

Question 64

What is the main drawback of point-based filters?

Answer 64

They cannot consider neighbouring pixel context and thus fail to remove spatial noise effectively.

Question 66

What is a neighbourhood (spatial) filter?

Answer 66

A filter that modifies a pixel based on the values of its surrounding pixels using a kernel.

Question 67

What is a box filter?

Answer 67

A linear filter where all weights are equal and the output is the average of neighbouring pixels.

Question 68

What is a Gaussian filter?

Answer 68

A linear filter where weights decrease with distance from the center, preserving edges better than a box filter.

Question 70

What is median filtering?

Answer 70

A non-linear filter that replaces each pixel with the median value in its neighbourhood.

Question 71

Why is median filtering effective for salt-and-pepper noise?

Answer 71

Because it removes outliers without averaging, thus preserving edges.

Question 72

What is the shape of a median filter window with radius 1?

Answer 72

3x3

Question 73

What is the main advantage of median filters over mean filters?

Answer 73

They preserve sharp edges while removing isolated noise.

Question 75

What is morphological filtering?

Answer 75

A set-theoretic approach that uses shape-based structuring elements to modify image structures.

Question 76

What is a structuring element in morphology?

Answer 76

A predefined shape used to probe the image, such as a square, line, or disk.

Question 77

What does erosion do in morphology?

Answer 77

It shrinks or thins bright objects in the image.

Question 78

What does dilation do in morphology?

Answer 78

It expands or thickens bright objects in the image.

Question 79

What is opening in morphological filtering?

Answer 79

Erosion followed by dilation; it removes small bright noise while preserving shape.

Question 80

What is closing in morphological filtering?

Answer 80

Dilation followed by erosion; it fills small holes and connects nearby bright regions.

Question 82

When would you use opening in image processing?

Answer 82

To clean up small bright noise elements without distorting larger shapes.

Question 83

When would you use closing in image processing?

Answer 83

To fill small dark gaps or holes within bright regions.

Question 85

What type of filter is convolution-based filtering?

Answer 85

Linear filtering

Question 86

What type of filter is median filtering?

Answer 86

Non-linear filtering

Question 87

What type of filter is morphological filtering?

Answer 87

Set-theoretic (non-linear, shape-based) filtering

Question 89

Why is dilation useful in object segmentation?

Answer 89

It helps to close gaps in segmented objects, improving connectivity.

Question 90

Why is erosion useful in image preprocessing?

Answer 90

It removes small noise or disconnects thin connections between regions.