MSc Dissertation Project Opportunities

Dr C P Willis and Dr D J Paddon

L-Systems for generating Virtual Plants

Virtual plants are computer models that recreate the structure and simulate the development of plants.

Virtual plant modelling combines mathematical formalism, biological knowledge, and computer graphics techniques.

An important modelling method is based on the theory of Lindenmeyer systems (L-systems). A fascinating aspect of this theory is the contrast between the relative simplicity of model specification and the apparent complexity, intricacy, and the visual realism of the resulting forms. L-systems allow us to:

• accurately recreate the structure and development of plants;

• show how the evolution of architectural parameters(branching angles, elongation rates, vigor of branches, etc.) affects the appearance of plants;

• simulate plant physiology and investigate the effects of manipulations(e.g. pruning) or different external conditions(local light microclimate, water availability, crowding) on plant development; and

• simulate plants not only in isolation, but growing and interacting with other plants.

Applications include:

• generating photorealistic virtual worlds

• tools for exploring desirable directions of breeding and manipulating ornamental plants for maximum visual impact.

• tools for simulating growth and aging in different environmental conditions

• teaching of botany and landscape design

Current research problems include:

• simulation and visualisation of interaction between plants and their environment.

• modelling of aging (particularly in trees)

• controlling the overall shape of modelled plants

• realistic modelling of complex ecosystems such as forests.

MSc projects could be undertaken in any of these areas, but the work for CM50175 would include a thorough investigation of the current state of research in all aspects of L-systems, in order to identify a specific project.

Recognition of solar features

Our sun displays a wealth of interesting surface features, the most observable of which are sunspots. Sunspots have been observed for hundreds of years — Kepler saw a sunspot in 1607 (though he attributed it to a transit of Mercury across the sun’s disk), but Chinese records of sunspots go back to 28BC.

Sunspots are generated by changes in the sun’s magnetic field. They grow, expand, shrink and die away, usually developing into complex clusters before dwindling back into nothing. Sunspot activity rises to a maximum every 11 years, and then falls away to a minimum before rising again - the last maximum was in 2000.

Image processing can be used to take white light images of the sun’s disk (these are widely available from sites such as NASA) and determine the boundaries of the sunspots (both umbras and penumbras, if present) and the division of the spots into related groups.

Problems with recognition of sunspots include:

• the complexity of many sunspots;

• determining whether a particular spot is “sufficiently close” to be part of a cluster;

• determining the classification of a particular sunspot or sunspot group.

Current research problems

• Improving automatic recognition of sunspots and their groupings from white light images of the sun.

• Classifying recognised sunspots using one of the existing visual classification systems such as the Modified Zurich Classification

An MSc project would probably concentrate on sunspot recognition and grouping, and extend into classification if feasible.

Dynamic Discontinuity Meshing

Shadows are extremely important features in a scene. They provide essential visual cues, hence if they are missing or misrepresented they can cause glaring visual anomalies. Discontinuity meshing is a technique that computes the shadows cast by objects, and provides the most complete description of a shadow because it is able to locate discontinuities within the penumbra.

The aim of discontinuity meshing is to generate high quality meshes that accurately encode the discontinuities across a surface. Unfortunately discontinuity meshing is computationally very expensive, because it attempts to determine all umbra and penumbra boundaries within a scene. If you consider the complexity of shadows that is cast in the MSc lab, by multiple slatted light sources, you can probably guess that the effort involved in meshing this lab would be considerable.

Current research problems include:

• Extremal discontinuity meshing.

An extremal discontinuity mesh contains only discontinuity lines that form the minimal and maximal shadow boundaries. The minimal boundary is defined by the boundary that separates the umbra and penumbra, and the maximal boundary is defined by the maximum penumbra boundary. By casting only an extremal discontinuity mesh the total number of discontinuity lines can be greatly reduced.

In scenes containing multiple light sources there will be interactions between the lights in the way that shadows are cast. Thus, there will be a consequent increase in the complexity of the meshing due to overlaps between the extremal meshes generated from each light source.

A possible project could build extremal meshes of scenes with multiple light sources, and use a technique such as perception metrics to reduce the complexity of the meshes.

• Accurate lighting models

Surprisingly little has been done in computer graphics to determine how we illuminate objects in a virtual reality scene to obtain photographic quality images. Photographers have a wide understanding of how light and shadow affects surfaces, and thus they know how to place lights, and what sort of lights to use in order to appropriately illuminate an object they wish to photograph.

Some work has been done on simulation of stage lighting for the theatre, but there is considerable scope for work on simulating specific types of lighting, including different types of light source (spots, diffusers etc).

A possible project would be to review the existing work and develop simulations that start to approximate the quality and effects that a photographer would expect.