Spatial filtering

Question 1

The kernel is a spacial filter

Answer 1

True
Local operations are defined based on a
filter/kernel.
The kernel defines–A neighborhood
A weight associated with eachpixel involved in the computation.
Local operations are performed in the spatial domain of the image
The kernel is AKA spatial filter

Question 2

convolution and correlation are equal

Answer 2

True
In the CV context, convolution and correlation are often used as synonyms– Usually, correlation is evaluated
* But it is called convolution!
* Filters are usually symmetric– Convolution and correlation are equal

Question 3

In convolution filter
the filter weights cannot change the image brightness if the sum of the weights is <1

Answer 3

False
The filter weights can change the image
brightness
* Brightness is unchanged if: the sum of weights of i = 1. where i belongs to I.

• This is obtained by a normalization factor

Question 4

which of the following are non spatial filters
1)average
2)derivative
3)min
4)single pixel operatiors

Answer 4

which of the following are non spatial filters
1)average
2)derivative
3)min
4)single pixel operators

A: all are spatial filters

Question 5

Consider the following statements about averaging filter

-The size of the filter can be
increased
-Larger filters: stronger smoothing

Answer 5

Consider the following statements about averaging filter

-The size of the filter can be
increased
-Larger filters: stronger smoothing

both are True

Question 6

averaging filters are non separable

2d convolution is possible in both vertical and horizontal manner

2d convolution filter doesn’t have to be a square

Answer 6

averaging filters are non separable- False

2d convolution is possible in both vertical and horizontal manner - true

2d convolution filter doesn’t have to be a square - true

A square filter of size 𝑛 × 𝑛 may be separated
in– A filter of size 𝑛 × 1– A filter of size 1 × 𝑛
* Separable filters: 𝑤
𝑥, 𝑦 →𝑤𝑥
𝑦– Can be applied on rows, then columns (or vv)– O(MN(a+b)) instead of O(MNab) – faster

Question 7

separating the averaging filter makes it faster

Answer 7

True

A square filter of size 𝑛 × 𝑛 may be separated
in– A filter of size 𝑛 × 1– A filter of size 1 × 𝑛
* Separable filters: 𝑤
𝑥, 𝑦 →𝑤𝑥
𝑦– Can be applied on rows, then columns (or vv)– O(MN(a+b)) instead of O(MNab) – faste

Question 8

Consider following statements about the first order derivative

1. Is zero in flat segments
2. Is zero on the onset of a step/ramp
3. Is zero along ramps

Answer 8

1. Is zero in flat segments - true
2. Is zero on the onset of a step/ramp-false
3. Is zero along ramps-true
4. Is zero in flat segments
5. Is non-zero on the onset of a step/ramp
6. Is non-zero along ramps

Question 9

Consider following statements about the second order derivative

1.Is zero in flat segments
2. Is non-zero on the onset and at the end
of a step/ramp
3. Is non - zero along ramps of constant slope

Answer 9

1.Is zero in flat segments -true
2. Is non-zero on the onset and at the end - true
of a step/ramp
3. Is non - zero along ramps of constant slope - false

1. Is zero in flat segments
2. Is non-zero on the onset and at the end
   of a step/ramp
3. Is zero along ramps of constant slope

Question 10

Consider following statements about the first order derivative

-they all have a scalar output
-they are all non linear
- they are non isotropic

Answer 10

Consider following statements about the first order derivative

there are 3 operators: gradient, gradient module, approx grad module

-they all have a scalar output - false: gradient has a 2d vector output
-they are all non linear- false - only gradient is linear
- they are non isotropic - false . only gradient module is non isotropic non ideally.

Question 11

The sum of elements in roberts and sobel is 0

Answer 11

The sum of elements in roberts and sobel is 0
true

Question 12

Sobel gradient can be used to get the boundary of a transparent lens

Answer 12

Sobel gradient can be used to get the boundary of a transparent lens
True

Question 13

1)Laplacian is a second order derivative
2)Laplacian is always subtracted
3)Laplacian is used for sharpening
4) it can enhance transitions in an image

Answer 13

1)Laplacian is a second order derivative :true
2)Laplacian is always subtracted : false;
Laplacian is subtracted if the center weight is negative
3)Laplacian is used for sharpening : true
4) it can enhance transitions in an image: true

Question 14

Linear filters suppress components

Answer 14

Linear filters suppress components:False

Non-linear filters can suppress components– Ex: a median filter can remove a spike (a single element that is strongly different from the others)– In our context: an element is a pixel

Question 15

A corrupted image may be restored using
a)averaging filter
b)sobel filter
c) non linear filters
d) derivative filters

Answer 15

A corrupted image may be restored using
a)averaging filter: true: is a convolution filter used for restoring and smoothening
b)sobel filter: false: sobel and roberts are first order derivatives; they sharpen the image; they are also convolutional filters. therefore linear.
c) non linear filters: true: min,max,median
d) derivative filters : false:it sharpens the image.

Question 16

Max filter: highlights pepper noise, removes salt noise
* Minfilter: highlights pepper noise, removes salt noise

Answer 16

Max filter: highlights pepper noise, removes salt noise - False
* Minfilter: highlights pepper noise, removes salt noise - True

Max filter: highlights salt noise, removes pepper noise
* Minfilter: highlights pepper noise, removes salt noise