Excercises

Exercises 179
that the least-squares approach works for this noise-free problem. And when we use the
synthetic data gf , which better represents the real measured data, then the reconstruction
is no longer perfect, but it better resembles the reconstructions that we compute from real
data coming from an object that is not pixelated on the given pixel grid.
Exercises
9.1. A Very Small System
We consider the CT problem with a 2 2 image and five parallel rays shown in
Figure 9.14, where the geometry is scaled such that the length of each ray through
one pixel is one.
Show that the corresponding system of linear equations A f b for the line
model takes the form

1 0 0 0 1
1 f1
0 1 0 3
f2
f3 5 .
0 1 1 0 (9.37)

0 1 0 1 7
f4
0 0 0 1 4
Can you guess the solution?
b1 1
b2 3
b3 5
f1 f3
b4 7
b5 4
f2 f4
Figure 9.14. A very small system; the length of each ray through a pixel is one.
180 Chapter 9. Discretization Models and the System Matrix
9.2. Setting Up a System Matrix

We use the function paralleltomo from the MATLAB software package AIR
Tools II to generate parallel-beam CT problems. In MATLAB you can use help
paralleltomo or doc paralleltomo to get information about how to use this
function.
Write A = paralleltomo(125,4:4:360) to generate a 15930 15625 system
matrix A for a problem with image size N N 125 125 and 90 projection
angles 4 , 8 , 12 , . . . , 360 . The default number of rays (detector elements) is
round 2N 177. Use spy(A) to display the nonzero structure of the matrix.
How dense is it?
Let ej denote the vector of all zeros, except for a one in element j. Convince
yourself that the product A ej is the jth column cj of the system matrix A. Hence,
cj is the Radon transform (the sinogram) of an image with a single nonzero pixel.
Show the sinograms for several values of j by means of the commands
ntheta = 90; % Number of angles

p = size(A,1)/ntheta; % Number of detector elements
imagesc( reshape(A(:,j),p,ntheta) ) % Sinogram
You should see discrete sinusoids—the discrete nature is due to the finite size of
the image and the detector elements.
Now recall that the ith row riT of A represent a discretization of the line inte-
gral that produces the data in the ith detector element; see (9.24). Perform the
same type of experiment as above, where you plot several rows of A as 125 125
images. You should see straight lines that correspond to the rays; the jagged ap-
pearance is again due to discretization issues.
9.3. Sinograms of Simple Images
Figure 9.15 shows the sinograms of two simple geometric objects on a zero back-
ground, generated with the same system matrix as above. Can you figure out what
these objects are just by looking at the sinograms?
Figure 9.15. Two sinograms of simple objects.

Exercises 181
Try to generate the corresponding images X and show the corresponding sinograms
S = reshape(A*X(:),p,ntheta). Can you reproduce the two sinograms?
9.4. A “Bad Pixel” in the Sinogram
Assume that we have a single bad pixel in the detector, e.g., a detector element that
always has intensity one (due to a fault in the CCD). Such a bad pixel produces a
horizontal line in the sinogram, because it is independent of the projection angle.
Let us study the influence of this bad pixel on the reconstruction. Generate a
sinogram S which is zero everywhere, except for the pixels in a horizontal line,
which all have intensity one. Then perform these two experiments with b = S(:).
Compute the back-projection A’*b and show it as an image.
Compute an FBP reconstruction and show it as an image; use the MATLAB
function iradon or the function fbp from AIR Tools II (or both).
In both cases you should see a distinct structure. This structure will appear as a
very annoying artifact when we compute a reconstruction from sinogram data with
a bad pixel.
9.5. Interpretation of Projection Followed by Back-Projection
In Chapter 6 we introduced the back-projection R7 (6.1) and explained it as a
smearing and summation process. Moreover, we saw that the process of applying
a forward-projection to an image f , followed by a back-projection, produces a
blurred version of f ; see Figure 6.8. This can easily be verified experimentally by
means of radon and iradon.
At this stage, where we have matrices A and AT that represent the forward- and
back-projections, we can illustrate this with an example. Recall that a black image
with a single white pixel is represented by a vector f rj s with zero elements except
for a single element xj 1 in position j. What happens when we forward-project
this image and then apply back-projection? The result is an image which we can
write as
f˜rj s AT A f rj s AT cj . (9.38)
This image shows the point spread function for pixel πj associated with the forward-/
back-projection process.
Choose an image size N and a vector theta of projection angles, generate the
system matrix A, compute and display x̃rj s as an image for different choices of j,
and comment on the results.

Excercises

Uploaded by

Copyright:

Available Formats

Excercises

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Excercises

Uploaded by

Copyright:

Available Formats

Exercises 179

Can you guess the solution?

9.2. Setting Up a System Matrix

ntheta = 90; % Number of angles

Figure 9.15. Two sinograms of simple objects.

You might also like