Proposed BSc/MSc Project for AY 2005-2006

 

Moments based super-resolution shape reconstruction algorithms for photogrammetry

 

Keywords: Photogrammetry, Object (Shape) determination from moments, Sampling signals with Finite Rate of Innovation (FRI), Complex-moments, Annihilating filter methods.

 

Required skills: Basic knowledge of signal processing, 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: Photogrammetry (or remote sensing) is the technique of measuring 2-D (or 3-D) objects (or their coordinates) from photographs, or from the imagery stored on electronic media through CCD cameras or radiation sensors such as optical or microwave scanners. Its most important feature is the fact that the objects are measured without being touched. Photogrammetry has many applications such as determining crop yield, illegal intrusion in vast land fields, military and defence surveillance, in geophysics, and in robot vision. Additionally, in many situations, these applications can be processed in such a way that the resultant imagery can be modelled with piecewise polynomial approximations.

 

In practice, due to hardware of the CCD camera (lenses), and in order to obtain efficient data storage, the obtained image is a prefilterd and sampled version (blurred and lower resolution) of the original (analogue) non-bandlimited scene. Hence it is not possible to precisely determine (reconstruct) desired 2-D shapes from that blurred (or sampled) version by applying Shannon’s sampling theories.  However, using recently developed sampling theories (that go beyond Shannon’s work and known as sampling of signals with Finite Rate of Innovation (FRI) ), one can precisely reconstruct the original shapes from the blurred version assuming that the shapes follow piecewise polynomial model. Furthermore, it is interesting to know that unknown objects (such shapes) can be determined from sets of their moments. For example zetoth order moment (averaging) of a 2-D shape gives its area. The ratio of first and zeroth order moments gives its centre of gravity, and a combination of its higher moments gives information about its boundary.

 

In this project, student will gain the understanding of recently developed sampling theories (FRI sampling), fundamental aspects of digital imaging, photogrammetry, various types of moments (geometric and complex) and their signal/image processing applications in determining image shapes. Student will start with estimation of simple shapes such as polygons, circles, and ellipses, and thereafter other smooth objects with unknown boundaries. The student is expected to understand and verify the existing algorithms, and contribute in optimising these algorithms by extracting minimum possible coordinates to represent such shapes from their finite moments. Finally, the student is expected to develop a professional looking photogrammetry GUI for visualization of the results in Matlab (or C++ or Java) for phantom images and simple real images.