ECE 4250 Assignment 4

ECE 4250
Assignment 4

General Instructions
The assignment submissions are via Canvas. There are two parts: a problem set and a coding
(programming) component. Your answers to the problems can be written up any way you want
(e.g., using Latex or scanning your handwritten answers). As long as we can read it, we will
grade it. For the programming part, you will code in Python 3, using Jupyter Notebook. Make
sure that all cell outputs are clearly visible before saving the notebook. Thus, we recommend
that you re-run your code after you are done editing. You should then print this file into a pdf.
For the coding component, please submit both the “.pdf” and “.ipynb” files. Please zip up all
your work into a single file and submit that.
Any acknowledgments of collaboration, answers to questions, comments to code or other notes
should all be included in your problem set answer sheet and/or Jupyter Notebook. General
guidelines regarding assignments apply. In other words, you can collaborate at the ideation/brainstorming stage, but whatever you submit should represent your own individual work. You are
not allowed to copy/paste others’ code/answers or work on same code with someone else. You
are not allowed to use functions outside of the Python Standard Library, unless specified otherwise or you have received permission to do so. Your code will be graded on readability,
executability, and accuracy. You are strongly advised to use detailed commenting that explains the algorithmic steps in your code. Explain every function and loop. Use piazza to ask
questions. Take advantage of TA and Instructor Office Hours to seek help.
Problem Set
[60 pts, 20 per question]
Question 1. • Two images x1(m, n) and x2(m, n) have unnormalized histograms h1 and
h2. Give a condition under which one can determine these image histograms from the
unnormalized histogram of x1 + x2. Describe the procedure you would implement to
compute the individual histograms.
• Give a single integer-valued intensity transformation for spreading the intensities of an
input image so lowest intensity is 0 and highest intensity is L − 1.
Question 2. Consider following 2D image:
cos(2πµom/M + 2πν0n/N),
where m, m0 ∈ [M] and n, n0 ∈ [N]. Derive the DFT.
1
Question 3. Consider a 3 × 3 spatial mask that averages the four closest neighbors of a pixel
(m, n), but excludes the pixel itself from the average.
• Express the filter as a 2D array (in spatial coordinates)
• Derive the 2D discrete Fourier transform of this filter
• Is this a high or low pass filter? Explain your answer.
Programming Questions
1 Histogram Equalization
[25 pts]
• Implement the basic histogram equalization algorithm. Recall that this method applies
following intensity transformation:
T(x) = (L − 1) Z x
0
px(w)dw,
where px is the normalized histogram of the image, and the intensity values are assumed
to be in [L]. Apply this algorithm to X.png and visualize result.
• Implement an improved (exact) histogram equalization algorithm based on the paper
discussed in lecture (Coltuc, Dinu, Philippe Bolon, and J-M. Chassery. ”Exact histogram
specification.” IEEE Transactions on Image Processing 15.5, 2016). Apply this algorithm
to X.png and compare the result to previous one.
2 Spatial Transformations
[25 pts]
For the following, the output image should be the same size as the input image. Outside of
boundary values can default to zero. You are allowed to use a readily available interpolation
function. Note that the “transformed image” might not have the field view to show the entire
original scene.
• Write a function called Rotate2D that accepts an input image and a rotation angle (in
degrees) and outputs an image that is the input rotated around the center pixel of the
image.
• Read the image X.png. Rotate it by 45 degrees, save this as RotatedX.png, and visualize
this image.
• Write a function called Translate2D that accepts an input image and a translation (shift,
displacement) vector (in 2D, pixel units); and outputs an image that is the input image
shifted/translated by the provided displacement vector.
• Shift X by [30, −70] pixels, save this as ShiftedX.png, and visualize this image.
2