|
|
4th Year Projects |
|
My work focuses on recovering and analysing the reflectance properties of surfaces. Reflectance characteristics are an intrinsic property of a surface and provide an invariant measure under various lighting conditions. Knowledge of reflectance may be used in a multitude of ways.
The 4th year projects I am proposing involve analysing recovered reflectance, in particular the Bidirectional Reflectance Distribution Function (BRDF) (just a fancy name for a function that defines the reflectance properties of a surface). The first two projects involve using reflectance information for recognition and classification and are application based, while the remaining two projects are more mathematically inclined and involve analysing arbitrary reflectance functions. These projects will most likely use previous work of mine for reflectance recovery. I will be readily available to provide assistance with this and can also help with some of the hands on project work. Ideally, each project could lead to the joint publication of an academic paper. If you are interested in any of the projects listed below or would just like to discuss possibilities, then don’t hesitate to drop me an e-mail or come by the Fallside lab sometime. Good luck with your project choice! |
| Project 1 | Material Recognition using Surface Reflectance Properties |
|
Often it is necessary to recognise similar objects of differing materials, when recycling for example. Reflectance is an intrinsic surface property and may be used to achieve this task. One could perform recognition either by analysing the recovered BRDF directly (using wavelet representations, principal component analysis (PCA) or fitting parametric models), or by using a database of models to determine the most similar reconstructed image. |
|
| Project 2 | Classification using Surface Reflectance Properties |
|
Reflectance properties may also be used for surface classification, for example determining skin cancer from healthy skin. Summary information could be extracted from recovered BRDFs by the possible techniques outline above, which could then be applied to a support vector machine (SVM) for classification. |
|
| Project 3 | Enforcing Constraints on Recovered BRDFs |
|
Physically realisable BRDFs must satisfy certain physical constraints. These include symmetry, known as Helmholtz reciprocity, and energy conservation. Often BRDFs are also isotropic. Due to noise and uncertainty, recovered arbitrary BRDFs may not exhibit these properties and so a more accurate representation may be achieved by enforcing them. This project would involve investigating various techniques for enforcing such properties, for both spatial and wavelet BRDF representations. Improvements may be analysed by examining both the refined BRDF and also novel images rendered using the refined BRDF. |
|
| Project 4 | Statistical Parametric BRDF Models |
|
Current parametric BRDF models are based on surface micro-geometries. Although these are very useful, huge assumptions are made on the surface geometry. An alternative approach to constructing parametric BRDF models could involve analysing the statistical properties of BRDFs, in particular the statistical nature of wavelet representations. This project would involve analysing BRDF statistics, proposing a parametric statistical model, and then fitting this model to various surfaces and evaluating its performance. |
|
| Last modified: August 2003 | Home |