Scribd
Upload
14 views3 pages

Ex1 Hints

This document provides programming hints for Coursework 1 in the COMP0114 Inverse Problems in Imaging course, focusing on Matlab and Python implementations. It covers techniques for solving underdetermined problems, singular value decomposition, and performing convolutions and Fourier transforms, including specific function usage and visualization tips. The hints are structured to guide students in using both programming environments effectively for their coursework tasks.

Uploaded by

Suvadeep Maiti
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
14 views3 pages

Ex1 Hints

This document provides programming hints for Coursework 1 in the COMP0114 Inverse Problems in Imaging course, focusing on Matlab and Python implementations. It covers techniques for solving underdetermined problems, singular value decomposition, and performing convolutions and Fourier transforms, including specific function usage and visualization tips. The hints are structured to guide students in using both programming environments effectively for their coursework tasks.

Uploaded by

Suvadeep Maiti
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

COMP0114 Inverse Problems in Imaging.

Coursework 1 Hints

Introduction
This document is designed as support for Coursework 1, containing programming hints for
Matlab and Python.

——————— HINTS ———————

1. Solving Underdetermined Problems

a.) Matlab: You can use the concept of an anonymous function or function handle and define
an anonymous function Phi(x,p) that computes Φ.
Python: You can use the concept of lambda functions or a function definition def to
compute Φ.
b.) Matlab: You can use the function fmincon to solve the constrained optimisation problem
here, study its documentation to figure out the correct orders and data types of input and
output arguments.
Python: The function scipy.optimize.minimize can be used to solve the above con-
strained optimization problem.

c.) Matlab: To plot the function and points, you need to study the MATLAB functions of
graphics. Familiarise yourself with Matlab’s 2D graphics (MATLAB > Graphics > 2-D
and 3-D Plots > Line Plots) to plot your results as shown in Figure 1: one may use hold on
to plot lines and points on the same figure and modify the configurations of plot functions
for customised shapes and colours.
To export the line plot, consult the documentation (MATLAB > Graphics > Printing
and Saving > Save Figure for Document or Presentation) and use the functions print or
saveas to save the graphic in a high quality image format (for instance as .pdf or .png;
avoid using compressed formats such as .jpeg). Clip the exported image if necessary (see
”Tools”).
Python: Familiarize yourself with the plotting library matplotlib to produce similar
high quality figures.
d.) Matlab: You can directly compute the pseudo inverse by pinv. There is another possi-
bility to compute a robust solution to a system Bx=b with the backslash operator B\b. Do
these methods yield the same solution?
Python: You can use the function numpy.linalg.pinv. Similarly to Matlab, the library
numpy also provided a class linalg which performs the backslash operation.
2. Singular Value Decomposition

a.) Matlab: You can also use the function linspace.


Python: The library numpy contains a function linspace for this purpose.
b.) Matlab: To visualise a vector g, simply plot it by: plot(g).
Python: you can simply use matplotlib.pyplot.plot(g) to visualizse a vector g.

d.) Matlab: One way of visualising A is to use MATLAB’s imagesc function to plot A and
and use saveas to export the plot as a .png. Often, it is advantageous to export images
directly into a raster graphic format. This can be done using Matlab’s imwrite function,
but requires a rescaling of A and the explicit definition of a colormap:
>> Aimg = ceil(A/max(A(:))*256);
>> colorMap = parula(256);
>> imwrite(Aimg,colorMap,’Aimage2.png’)
Consult the documentation to comprehend the above commands.
Python: We require the library opencv (For installation use for instance
pip install opencv-python) to obtain similar results as in Matlab:

import cv2
import numpy as np
Atmp = np.array(np.ceil(A/np.max(A)*256), dtype = np.uint8)
Aimg = cv2.applyColorMap(Atmp, cv2.COLORMAP_JET)
cv2.imwrite("Aimage3.png",Aimg)

e.) Matlab: Use the function svd to compute the SVD of A.


You can check if two matrices A and B are equal (or close) by computing the norm of their
difference: norm(A-B).
Python: Use numpy.linalg.svd to compute the SVD of A.
You can use numpy.linalg.norm to compute the norm.

f.) Matlab: One can construct the pseudoinverse W † of W as a sparse matrix by using the
spdiags function (familiarise yourself with the concept of sparse matrices), in order to
save space for large dimensional cases. One can use the function spy to check whether W †
has the desired form and verify that A† = V W † U T by using Matlab’s pinv function to
compute A† .
To plot diag(W ) on a logarithmic scale on the y-axis, you can use the function semilogy
instead of plot, i.e. semilogy(diag(W)).
Python: Construct the pseudoinverse W † of W as a sparse matrix by using
scipy.sparse.spdiags function. Once can use function matplotlib.pyplot.spy to
check the form of W † and verify that A† = V W † U T by function numpy.linalg.pinv.
You can use matplotlib.pyplot.semilogy(numpy.diag((W)) to make a plot with loga-
rithmic scaling on the y-axis.

3. Convolutions and Fourier transform


e.) As the output of the Fourier transform is complex, you have to look at real, imaginary or
absolute values separately for plotting purposes. It can be also useful to look at the output
with and without using shifting as explained in the following:
Matlab: For computational reasons FFT performs a shifting, to keep the correct order
of your signal x perform the Fourier transform by Fx=fftshift(fft(fftshift(x))) and
similarly for the inverse Fourier transform fftshift(ifft(fftshift(Fx))). For pointwise
multiplication in Fourier space use the operation: .*
Python: We require numpy.fft module to perform FFT as in Matlab:

import numpy as np
Fx = np.fft.fftshift(np.fft.fft(np.fft.fftshift(x)))
x = np.real(np.fft.fftshift(np.fft.ifft(np.fft.fftshift(Fx))))

You might also like