Proposed BSc/MSc Project for AY 2005-2006
Radon
transform based super-resolution algorithm for
shape reconstruction
Keywords: Radon transforms, Tomographic reconstructions, Geometric
moments, Sampling of signals with Finite Rate of Innovation (FRI), and Annihilating
filter methods.
Required skills: Basic knowledge of DSP, interest
(and aptitude) towards GUI development in Matlab (or C++ or Java).
Number of students: One.
Secondary support: Mr Pancham Shukla (RA and PhD student of Dr P L Dragotti)
Room 808, http://www.commsp.ee.ic.ac.uk/~pancham/index.html
Abstract: In many applications such as determination of crop yield,
illegal intrusion in vast land fields, military and defence surveillance,
geophysical explorations, and robot vision, it is important to determine 2-D
(or 3-D) objects (or their coordinates) with high precision from blurred and
sampled images (digital images) obtained trough CCD
cameras or radiation sensors (optical or microwave scanners). In many
situations, the actual imagery can be modelled by a set of shapes with Finite
Rate of Innovation (FRI) (e.g. polygons, circles, and ellipses). Super-resolution
shape reconstruction deals with precise determination of such shapes from their
blurred version(s).
It is interesting to know that shapes
and dimensions of unknown 2-D (or 3-D) objects can be determined
(reconstructed) by taking Radon transform projections at finite number of viewing
angles (within [0,180) degrees) and then projecting them back on the same
plane. The famous example of this technique is tomographic reconstruction (CT
scanning) of human body. Furthermore, the Radon transform projections of an object
are also linked with the object’s geometric moments. Using this link, one can
take moments of the projections of the (blurred and sampled version of the)
given FRI object at various angles, and then by using sophisticated back
projection algorithms (e.g. annihilation filter for moments), one can determine
the coordinates of the object with infinite precision.
In this project, student will gain
the understanding of recently developed sampling theories (FRI sampling),
fundamental aspects of tomographic reconstruction used in modern medical
scanners, connection between moments and Radon projections, and its signal/image
processing applications in determining image shapes. Student will start with
estimation of simple shapes such as polygons and thereafter polygons containing
polygonal holes, and other smooth objects with unknown boundaries. Finally, the
student is expected to develop a professional looking shape reconstruction GUI in
Matlab (or C++ or Java) for phantom as well as real images. The student will
acquire a privilege of using high-end digital camera of the laboratory for his
GUI development.