Wolfson College, Oxford: Advanced Computational Science and Engineering

As an extension to the Centre of Excellence activities a joint affiliate centre between CCSE and Wolfson College of Oxford University was set up in Oxford with the state of the art clustercomputing facilities and with two full time researcher concentrating on Advanced Computational Science and Engineering (ACSE). The affiliate centre hosts every year 2-3 visiting scholars from CCSE to interact with Oxford scientist for further researcher training.

The research of ACSE is collaborative effort with scientists in Theoretical Physics (Professors Sir Roger Elliott, Douglas Abraham and Robin Stinchcombe) of the Department of Physics, in Information Engineering (Professor Mike Brady) of the Department of Engineering Sciences, in Materials Sience Department (Professor: Adrian Sutton), and in Mathematical Biology (Professor Philip Maini) of Mathematics Institute of the University of Oxford.

Robust segmentation of textured images

Researchers: Veit Schenk1 2
Department of Engineering Science, University of Oxford
Laboratory of Computational Engineering, Helsinki University of Technology

The aim of this project is to robustly segment, classify and quantify images of underwater scenes. The objects of interest are coral which are photographed at regular intervals. This allows to monitor the change of the coral over time, thus providing information of how environmental factors affect the growth of coral. The difficulty with performing this task manually is that the number of individual corals is large and the measuring process consequently firstly very time consuming and secondly difficult to perform accurately and repeatedly. We propose the following system to (semi-)automate the process: The first step involves distinguishing coral from other objects such as sand, rock, algae etc. To achieve this, we use a boostingapproach to positively identify individuals of each species of coral of interest. This not only removes any objects which are not coral (or at least not of the species we are interested in), but also provides a starting point for the second stage, the quantification: we fit a deformable model of the species of interest to the individual candidates. Since the data is rather noisy, as illustrated in Fig. 52 and hence difficult to segment using plain edge-detectors, we use textons to jointly segment and classify individual regions of each individual coral. Fig.53 illustrates one particular species.

Figure 52

Figure 52: Section of a coral-colony. This illustrates the difficult nature of the images: the illumination is non-uniform, they are relatively noisy(in terms of image content) meaning that traditional edge-detectors would not work very well at identifying individual corals.

Figure 53

Figure 53: Closeup of a coral (Diploastrea heliopora). The centre is a very different texture compared to the ’outside’ which spreads radially out. The ourside is the same texture all around, but varies in orientation and scale.

The centre consists of a unique texture (small whitish-grey bubbles). The surrounding main part of the body consists of rings radiating out from this centre. The difficulty arising in this particular case is that the main body is strongly isotropic, i.e. standard region-based textons will produce different responses around the body, despite the entire area clearly consisting of the same texture. A further problem is the change of illumination due to the lighting coming from one side and the corals being raised, i.e. casting shadows on one side. We address the lighting-issue by using local energy operators, and extracting phase-information which is contrast/lighting invariant. The rotation-invariance issue is still an open problem. These textured regions are then used to determine the boundaries which are used to control the deformable model (consisting of ’energy’ terms, one related to the inherent deformation, the other one to the fit to the data). Once the deformable model has been fitted, information about the size of the coral can be extracted. (in each image, a ruler will be photographed as part of the scene in order to provide scale information). A future application will be to match a scene containing coral from one time-instant to another, thus providing information about the change over time. Initially, this comparison is performed manually.

Wetting effects at a grain boundary

Researchers: D. B. Abraham1, Ville Mustonen1 2 and A. J. Wood1
Department of Physics-Theoretical Physics, University of Oxford
Laboratory of Computational Engineering, Helsinki University of Technology

We study the statics and dynamics of a two-dimensional Ising lattice system with a grain boundary in the middle. The grain boundary can be introduced by weakening the vertical bonds of the lattice by a factor . By imposing Dobrushin boundary condition and keeping the system below bulk critical temperatures, an interface is formed (see figure 54). With the grain boundary there are two competing configurations for the interface to adopt; a straight interface with an angle and a dog-leg configuration which consists of two fluctuating sections of interface which are connected along the centre line by a further interface pinned to the defect line, see figure 55. We demonstrate that the crossover between these two is manifested by a phase transition.

Furthermore we studied the relaxation of the system using continuous time Monte Carlo with Kawasaki dynamics. The system without a grain boundary relaxes using capillary fluctuations, whereas in the case of a grain boundary mass transport through defect line dominates the relaxation process. This means that by introducing a defect line in to the system one is able to confine the mass transport to the minimum energy pathway. To draw an analogy to the precursor film phenomenon in the dynamics of wetting, one can say that the grain boundary acts effectively as a substrate with a chemical potential favouring spreading.

Figure 54

Figure 54: System without a grain boundary has an interface crossing the system in angle .

Figure 55

Figure 55: System with a grain boundary adopts a dog-leg configuration below the transition temperature.