Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: daixieit

ENGN4528 Final Exam

S1 2021

College of Engineering and Computer Science

Mid-semester Quiz, First Semester 2021

ENGN4528 Computer Vision

Question Booklet

Reading time: 15 minutes

Writing time: 1 hour

Uploading time: 15 minutes

There are 4 questions in total.

(Q1-Q4)

Please name your submission as

ENGN4528_quiz_u1234567.docx

(or PDF as you prefer)

Q1: (10 marks) [basic concepts]

Answer the following questions concisely. Each of the questions must be answered in no more than 5 lines of text. Longer answers will be penalized.

1.  Consider the HSV colour space. What does H stand for? [1 mark]

2.  Describe the information that the H channel contains. [1 mark]

3.  Is a Bilateral filter a linear filter? Explain in one or two sentences why or why not? [2 marks]

4.  Suppose that we have a histogram modification function as follows:

M1(v) = c(v/255)^,, where 0 <= v <= 255 (assuming pixel values for the camera are in the range of [0, 255], where ,= 0.3, and c is 1.

(a) Suppose that we apply this to the following image. Describe what the effect will be on the modified histogram? [2 marks]

(b) Suppose we now use a different function that operates as follows: M2(v) = c(v/255)^,, where v < 55

M2(v) = c((v-20)/255)^,, where 55 <= v < 75

M2(v) = c(v/255)^,, where v >=75

where,= 0.3, and c is 1.

What would be the new value for a pixel when its current value is 56? [1 mark]

(c) Describe in one or two sentences what the differences are between the

resultant images after applying M1 and M2 to the original one, respectively. [3 marks]

Answer the following questions concisely. If you are unsure about some part along the way, state your best assumption and use it for the remaining parts. Similarly, if  you think some aspect is ambiguous, state your assumption and write the answer as clearly as you can.

Q2: (10 marks) [Calculations]

1. Homogeneous transformations of a 2D image can be represented by a single 3x3 matrix. What would be 3x3 the matrix of transformation matrix that

results in an output image being translated 3 pixels to the left and 6 pixels

upwards? (where up and left are defined from an image origin at the top left- hand corner. [3 mark] Notes: detailed image coordinate system and the result after translation are defined below.

2. If you were to apply the following filter as a correlation:

to the blue and red regions separately in the following image patch:

Calculate the result of the correlation applied to the red (upper left) and the blue

square (bottom right) (the result of the correlation is one value for each square). [3 marks]

Describe what operation this filter performs/achieves in one or two sentences. [2 marks]

3. If you apply a filter F =

Followed by a filter G =

to an image. What is the equivalent 3x3 filter to these operations? [2 marks]

Q3. Basic Algorithm [8 marks]

(a) Briefly explain the steps of the RANSAC algorithm, be sure to define what an inlier is, and how inliers are used in the RANSAC algorithm. [4 marks]

(b) Briefly explain the steps of the histogram equalization algorithm. [3 marks]

(c) What is the effect of applying histogram equalization to animage twice? Justify your answer. [1 mark]

Q4. Algorithm Design. [12 marks]

Please design an algorithm to detect ellipses from the following binary image. Please list the key steps and ideas to detect ellipses. [Namely, we would like to find the

membership of each pixel whether it belongs to an ellipse or not and which ellipse it belongs to.] Note that the question must be answered in less than one page. Longer   answers will be penalized.

Notes:

The parametric ellipse equation is defined below

Where (x0 , y0) defines the centre of the ellipse, (a, b) defines the length of semi- major axes respectively.