Abstract— The paper considers a new iterative algorithm for
tomographic image reconstruction when a very limited number
of projections are registered. The algorithm uses a priori
information on the discrete values of the object function to be
reconstructed and is based on the known multiplicative
algebraic reconstruction technique (MART). The new
algorithm we called MART-AP organizes a cycle of “external”
iterations where the image is corrected at each iteration, using
the MART and an image mask synthesized with the help of a
priori information. It has been shown in numerical experiment
that the algorithm proposed helps completely remove streak-
like artifacts usually present on the tomograms which were
reconstructed in conditions of strongly incomplete data.
I. INTRODUCTION
HE problem of image reconstruction with strongly
incomplete data, for example, when a few (ten or less)
projections are only registered, arises in many areas of
practical tomography such as industrial nondestructive
testing [1], plasma emission tomography [2], tomography of
explosion-compressed metal shells [3], tomography of
hydrodynamic object [4] etc. Images from few-view
computed tomography are usually reconstructed with an
algebraic approach based on the derivation of a discrete
reconstruction model with which the problem is reduced to
the solution of an underdetermined system of algebraic
equations [2], [3], [5]. As a rule, the system is inverted with
algorithms that optimize its solution with respect to an
additional criterion. These are, for example, the
multiplicative algebraic reconstruction technique (MART)
[6] or the maximum entropy technique (MENT) [7]. But
even these techniques are incapable of removing the few-
view artifacts if the object has high-contrast low-frequency
structures. The artifacts are usually seen as streaks tangent to
the boundaries of the structures. Figure 1 demonstrates how
the artifacts add noise to the image and does not allow the
real structure to be recognized. Figure 1a shows a 2-D test
object which is described in section IV, and figure 1b
illustrates the result of its MART reconstruction from 9
model projections. The image of figure 1a has only two
colors: black and white. And the tomogram of figure 1b is
presented as a gray level image. Our numerical experiment
assumed sources to be arranged in a 150°-sector at equal
angular steps.
Manuscript received October 24, 2011.
V. V. Vlasov, A. B. Konovalov, and A. S. Uglov are with the Russian
Federal Nuclear Center – Zababakhin Institute of Applied Physics, PO Box
245, Snezhinsk Chelyabinsk Region, 456770 Russia (phone:
+73514654639; fax: +73514652233; e-mail: a_konov@mail.vega-int.ru).
The problem of streak artifact correction is considered in
many papers. Most authors come through using either
tomogram post-processing [8], or projection pre-processing
[9]. Only a few try to modify the reconstruction techniques
themselves. So, the authors of [10] proposed an original
modification based on the solution of an artifact recognition
problem, after which corrections to the cells in the space
where artifacts were detected were applied in another way
than to the cells in the space which was free of artifacts.
Unfortunately, the authors were a success only partly.
It becomes possible to improve a tomogram if we know a
priori, for example, the discrete values of the object function
to be reconstructed [11], [12]. What is of great importance in
this case is the way in which we incorporate this information
in the algorithm. For example, the threshold segmentation
usually used for tomogram post-processing is of low effect.
This is demonstrated in figures 1c and 1d. So, what is shown
in figure 1c is a result obtained in the segmentation of the
tomogram shown in figure 1b, where the threshold value
was equal to the halved maximum of intensity, i.e., with no a
priori information. And the image shown in figure 1d was
obtained in the segmentation with an a priori known
threshold. We must admit both unsatisfactory. Until recently
An a Priori Information Based Algorithm for Artifact Preventive
Reconstruction in Few-View Computed Tomography
Vitaly V. Vlasov, Alexander B. Konovalov, and Alexander S. Uglov
T
(a) (b)
(c) (d)
Fig. 1. A model test object (a), the result of its reconstruction with
MART (b), and the results of tomogram segmentation without (c) and
with (d) an a priori known threshold.
Proceedings of the 5th International Symposium on Communications, Control and Signal Processing,
ISCCSP 2012, Rome, Italy, 2-4 May 2012
978-1-4673-0276-0/12/$31.00 ©2012 IEEE