ECE 438 Image Analysis & Computer Vision Sample Test #2 NAME____KEY_____________

Answer all questions in space provided. Use back of the pages for extra work. Note that each question

is not weighted equally. You may use: 1) 1 sheet, 1-side of handwritten notes,

2) your own calculator. Show all your work. You should have 6 pages. You have 75 minutes.

#1) Mark the following statements True (T) or False (F)

___T__ Image analysis can be performed in both the spatial and spectral domains

___F__ Multiresolution is the best image segmentation method

___F__ In image analysis, application-specific feedback is of minor importance

___F__ With the Fourier transform, the phase contains information about the image contrast

___F__ Nearest neighbor classification is less computationally intensive then nearest centroid

___F__ Minimizing the number of mis-classifications is the only consideration when designing a

classifier

___T__ We often log-remap a Fourier spectrum due to the response of the human visual system

___F__ When developing a classification algorithm it is a good idea to test it with the same samples

used for developing it.

__T___ During image segmentation we look for objects that are homogeneous

__T___ The modulation and periodicity properties of the Fourier transform allow us to shift the origin

to the image center

__T___ When using a neural network it is best to preprocess the feature data with a PCT

__T___ Euclidean and city block distance metrics are special cases of the Minkowski metric

__F___ Edges in images consist of primarily low frequency information

__F___ Frequency domain coefficients have a one-to-one correspondence with the image elements,

I(r,c)

__T___ Split and merge segmentation requires use if a homogeneity test

#2) Giving the following basis images; a) Find: T(u,v), b) Verify that your answer to a) is correct by

finding I(r,c) from T(u,v) (write out the all the steps)

















1 1













1- 1

and I(r,c) =













6 4

4 2













1- 1-

1 1













1 1-

1- 1















416

b) I(r,c) =

;































2416

168

0441604416

044164416

need to normalize,

multiply by ¼

#3) a) Find the 5x5 Laws texture energy mask for spots and edges, b) Find the 5x5 Laws texture energy

mask for gray level and ripples c) Find the 5x5 Laws texture energy mask for ripples and waves d)

What, if any, preprocessing is necessary to use the Laws energy masks?

Spot = (-1,0,2,0,-1)

Edges = ( -1,-2,0,2,1)

Partial preview of the text

Download Sample Exam 2 with Answer Key - Image Analysis and Computer Vision | ECE 438 and more Exams Electrical and Electronics Engineering in PDF only on Docsity!

ECE 438 Image Analysis & Computer Vision Sample Test #2 NAMEKEY_________

Answer all questions in space provided. Use back of the pages for extra work. Note that each question

is not weighted equally. You may use: 1) 1 sheet, 1-side of handwritten notes,

2) your own calculator. Show all your work. You should have 6 pages. You have 75 minutes.

#1) Mark the following statements True (T) or False (F)

_T Image analysis can be performed in both the spatial and spectral domains

_F Multiresolution is the best image segmentation method

_F In image analysis, application-specific feedback is of minor importance

_F With the Fourier transform, the phase contains information about the image contrast

_F Nearest neighbor classification is less computationally intensive then nearest centroid

_F Minimizing the number of mis-classifications is the only consideration when designing a

classifier

_T We often log-remap a Fourier spectrum due to the response of the human visual system

_F When developing a classification algorithm it is a good idea to test it with the same samples

used for developing it.

T_ During image segmentation we look for objects that are homogeneous

T_ The modulation and periodicity properties of the Fourier transform allow us to shift the origin

to the image center

T_ When using a neural network it is best to preprocess the feature data with a PCT

T_ Euclidean and city block distance metrics are special cases of the Minkowski metric

F_ Edges in images consist of primarily low frequency information

F_ Frequency domain coefficients have a one-to-one correspondence with the image elements,

I(r,c)

T_ Split and merge segmentation requires use if a homogeneity test

#2) Giving the following basis images; a) Find: T(u,v), b) Verify that your answer to a) is correct by

finding I(r,c) from T(u,v) (write out the all the steps)

v 

u  

and I(r,c) = 

a)

b) I(r,c) = ;

need to normalize,

multiply by ¼

#3) a) Find the 5x5 Laws texture energy mask for spots and edges, b) Find the 5x5 Laws texture energy

mask for gray level and ripples c) Find the 5x5 Laws texture energy mask for ripples and waves d)

What, if any, preprocessing is necessary to use the Laws energy masks?

Spot = (-1,0,2,0,-1)

Edges = ( -1,-2,0,2,1)

a) Vector outer product =

b) Vector outer product =

c) Vector outer product =

Note: error in book for wave, Should be: [-1 2 0 -2 1]

d) Preprocess the image to remove artifacts caused by uneven lighting. This can be done by subtracting

the local average in a moving window.

#4) Given the following two features vectors, find the following distance and similarity metrics:

F

a) Euclidean distance=2.5, b) city block distance=4, c) maximum value=2, d) Minkowski distance, with

r = 2,=2.5 e) vector inner product=112, f) Tanimoto metric=0.

#5) Sketch a bimodal histogram, label the axes. Draw a line to threshold the corresponding image.

Briefly describe a method to automatically find the threshold.

This method is called minimizing within group variance , or the Otsu method :

Find the value of the threshold t that will minimize the within group variance, using equations in

4.3.3. This can done calculating the values for σ

w(t) for each possible gray level value and

After Erosion:

After Dilation:

#10) Given the following feature vectors, with two classes:

Class 1:

F

1 2 3 Class 2:^

F

1 2 3

Sample Exam 2 with Answer Key - Image Analysis and Computer Vision | ECE 438, Exams of Electrical and Electronics Engineering

Related documents

Partial preview of the text

Download Sample Exam 2 with Answer Key - Image Analysis and Computer Vision | ECE 438 and more Exams Electrical and Electronics Engineering in PDF only on Docsity!

ECE 438 Image Analysis & Computer Vision Sample Test #2 NAME____KEY_____________

Answer all questions in space provided. Use back of the pages for extra work. Note that each question

is not weighted equally. You may use: 1) 1 sheet, 1-side of handwritten notes,

2) your own calculator. Show all your work. You should have 6 pages. You have 75 minutes.

#1) Mark the following statements True (T) or False (F)

_T Image analysis can be performed in both the spatial and spectral domains

_F Multiresolution is the best image segmentation method

_F In image analysis, application-specific feedback is of minor importance

_F With the Fourier transform, the phase contains information about the image contrast

_F Nearest neighbor classification is less computationally intensive then nearest centroid

_F Minimizing the number of mis-classifications is the only consideration when designing a

classifier

_T We often log-remap a Fourier spectrum due to the response of the human visual system

_F When developing a classification algorithm it is a good idea to test it with the same samples

used for developing it.

T_ During image segmentation we look for objects that are homogeneous

T_ The modulation and periodicity properties of the Fourier transform allow us to shift the origin

to the image center

T_ When using a neural network it is best to preprocess the feature data with a PCT

T_ Euclidean and city block distance metrics are special cases of the Minkowski metric

F_ Edges in images consist of primarily low frequency information

F_ Frequency domain coefficients have a one-to-one correspondence with the image elements,

I(r,c)

T_ Split and merge segmentation requires use if a homogeneity test

#2) Giving the following basis images; a) Find: T(u,v), b) Verify that your answer to a) is correct by

finding I(r,c) from T(u,v) (write out the all the steps)

v 

u  

and I(r,c) = 

a)

b) I(r,c) = ;

need to normalize,

multiply by ¼

#3) a) Find the 5x5 Laws texture energy mask for spots and edges, b) Find the 5x5 Laws texture energy

mask for gray level and ripples c) Find the 5x5 Laws texture energy mask for ripples and waves d)

What, if any, preprocessing is necessary to use the Laws energy masks?

Spot = (-1,0,2,0,-1)

Edges = ( -1,-2,0,2,1)

a) Vector outer product =

b) Vector outer product =

c) Vector outer product =

Note: error in book for wave, Should be: [-1 2 0 -2 1]

d) Preprocess the image to remove artifacts caused by uneven lighting. This can be done by subtracting

the local average in a moving window.

#4) Given the following two features vectors, find the following distance and similarity metrics:

F

F

a) Euclidean distance=2.5, b) city block distance=4, c) maximum value=2, d) Minkowski distance, with

r = 2,=2.5 e) vector inner product=112, f) Tanimoto metric=0.

#5) Sketch a bimodal histogram, label the axes. Draw a line to threshold the corresponding image.

Briefly describe a method to automatically find the threshold.

This method is called minimizing within group variance , or the Otsu method :

Find the value of the threshold t that will minimize the within group variance, using equations in

4.3.3. This can done calculating the values for σ

w(t) for each possible gray level value and

After Erosion:

After Dilation:

#10) Given the following feature vectors, with two classes:

Class 1:

F

F

F

1 2 3 Class 2:^

F

F

F

a) Using the Nearest Neighbor classification method, and the absolute value distance metric, classify the

following unknown sample vector as Class 1 or Class 2:

F

Class 1: d1= 6, d2 = 11, d3 = 10; Class 2: d1=7, d2= 2, d3=

Smallest distance is 2, therefore Class 2.

b) Use K Nearest Neighbor, with K = 3

The closest 3 have distances of 2 (class2), 6 (class2) and 6 (class1), therefore answer is Class 2.

#11) Given the following, what will be the resultant pixel values after operating on the following

image? Assume all rotations of the surrounds are included in S.

IMAGE

ECE 438 Image Analysis & Computer Vision Sample Test #2 NAMEKEY_________