PhD Projects

The following list of projects gives you an idea of the type of research that we do at Computing at the University of Dundee. However, your own ideas are welcome and it may be possible to work in other areas of research.

Research is divided into the following topics:

Please contact a potential supervisor if you are interested in any of the projects, or have your own idea for a project.

Applied Computing & HCI

Augmentative and Alternative Communication (AAC) systems for non-speaking people

Non-speaking people who use Augmentative and Alternative Communication (AAC) systems typically have low rates of communication which reduces their ability to interact with others. Research and development continues in the quest to improve the effectiveness of AAC systems in terms of communication rate and impact. One strategy involves making the basic unit of communication an entire utterance, and designing the AAC system to make the storage, retrieval and production of utterances as easy and efficient as possible. Some approaches take this further and include texts, narratives and/or multimedia material for use in conversation. AAC systems operating in such a manner require a structure for containing and managing conversational material and supporting the production of output during conversation. Ideally such a structure should be modelled on the way actual conversations proceed. A number of partial models for this have been presented thus far. This proposal would be to investigate and develop an integrated model that includes both the structure of a conversation and the way in which an AAC system might produce conversational output (e.g. utterances, texts, multimedia items or combinations of these).

Supervisor: Professor John Arnott


High Quality Speech Synthesis based on magnetic resonance imaging

Articulatory synthesis is a method of speech synthesis which artificially generates the acoustics produced by a virtual vocal tract. A 2-D version of this has been successfully demonstrated using a simple computer-generated vocal tract model. To produce high-quality synthesis, it is envisaged that a high-resolution 3-D model is required, and the obvious way to produce such a model is from Magnetic Resonance Image (MRI) scans of a human subject. The process that would be required is to take a series of MRI slice images of the head (from the larynx upwards), then within each slice, use vision processing techniques to identify the edges of the vocal tract (i.e. the interface between air and tissue. The location of the edges in each slice can then be extrapolated to produce the 3-D shape of the vocal tract. It would be useful to visualise this 3-D shape prior to using it for sound production. The challenges of this project are:

  • background research into MR imaging of the vocal tract area and articulatory synthesis
  • ethics pertaining to acquisition of MR images for this purpose
  • vision processing to locate the edges of the vocal tract in the image slices
  • image processing to synthesise a 3-D model shape from the boundary location and slice lo-cation data
  • signal processing to artificially generate sound using the 3-D model as the resonant chamber

Supervisor: Dr Iain Murray


HCI Challenges in Augmentative and Alternative Communication

Augmentative and alternative communication (AAC) attempts to augment natural speech, or to provide alternative ways to communicate for people with limited or no speech. Technology has played an increasing role in AAC. At the most simplest level, people who use AAC can cause a prestored message to be spoken by activating a single switch . At the most sophisticated level, literate users can generate novel text.

Although some children who use AAC become effective communicators, many do not - they tend to be passive communicators, responding mainly to questions or prompts at a one or two word level. Conversational skills such as initiation, elaboration and storytelling are seldom observed.

One reason for the reduced levels of communicative ability is that AAC technology provides the user with a purely physical link to speech output. The user is required to have sufficient language abilities and physical stamina to translate what they want to say into the code sequence of operations needed to produce the desired output.

Instead of placing the cognitive load on the user, AAC devices can be designed to support the cognitive and language needs of individuals, taking into account the need to scaffold communication as children develop into adulthood.

An on-going challenge when designing AAC systems is the increasing complexity of the user interface. This PhD topic will address the cognitive and perceptual challenges which arise when designing communication systems for children and adults who use AAC.

Supervisor: Professor Annalu Waller


Natural Language Processing and Personal Narrative

Relating one’s own stories in a conversational context is an important part of interactive communication. We express our individuality by remembering, formulating and retelling our own stories. This can and does pose significant problems to disabled, non-speaking people in terms of retrieving appropriate stories in an effective way during interactive conversation.

The “How Was School Today?” is a pilot study to develop a computer tool which helps children who cannot speak create a story about their day at school. Storytelling is an essential aspect of social interaction, and storytelling skills are developed through practice. It is difficult for non-speaking children to get such practice, our tool will help them. More specifically, we want to use various kinds of sensors to acquire information about where the child went, what she did, and who she interacted with; write a computer program which automatically creates a draft story based on this data; and create a story editing and narration interface which lets children edit the draft story and then tell it when they are happy with it. Possible sensors include GPS for tracking where children go, RFID tags for tracking what objects children interact with and hence their activities; and barcode scanners for recording with whom children interact.

This research area provides several opportunities for PhD projects:

Project 1: We have built a simple prototype system and have undertaken a small-scale evaluation with two children. We now wish to extend the project to allow users to generate more content. This will involve the use of natural language generation to produce different parts of a narrative, e.g. phrases such as “and then we”, “guess what happened next?” A PhD opportunity exists to investigate the how natural language generation can support language generation.

Project 2: The prototype only works in a constrained environment. A PhD opportunity exists to explore the use of context in conjunction with GPS and other sensor input can extend the environment in which the system is used.

Supervisor: Professor Annalu Waller


Interfaces for Phonic Based Communication and Mobility

Individuals with physical disabilities often use electric powered wheelchairs (EPWs) for mobility and Speech Generating Devices (SGDs). Both devices usually each have their own interface which means that the user needs to switch interfaces when changing from driving the wheelchair to communicating. The most common interface for controlling EPWs is a joystick.

Learning to read and write is a complex process which relies heavily on phonetic (sound) awareness. Nonspeaking children find this learning process problematic and seldom acquire literacy. A novel device (joystick) which allows direct access to spoken sound is currently being prototyped. A PhD opportunity exists to explore the relationship between mobility and language development.

Supervisor: Professor Annalu Waller


Maths and Computing

Subgraphs with simple structure

In many applications of graphs, good algorithms can be found in cases when the graph has a simple structure. If this is not the case, then a good alternative is to find a subgraph (as large as possible) which has a simple structure. The concept of fragmentability is one way of measuring this; in this case simple structure means components of bounded size. Work on this has mainly con-centrated on input graphs of bounded degree, for example it is known that for graphs of maximum degree three, there is a subgraph with small components containing three-quarters of the vertices, and the fraction three-quarters cannot be made larger in general.

There are several possible directions for further work: one is further work on fragmentability look-ing at other classes of graph, or trying to improve existing results, another is looking at directed graphs, and a third is using other meanings of "simple structure".

Supervisor: Dr Keith Edwards


Detachments of directed graphs

In graph theory, a detachment of a graph is a graph obtained by splitting vertices of the original graph into two or more subvertices and sharing out the incident edges among the subverti-ces. In a sense this is one of the oldest concepts in graph theory since an Eulerian trail can be viewed as a detachment of a graph into a cycle. Although quite a lot of work has been done on de-tachments of undirected graphs, the concept also applies perfectly well to directed graphs, on which very little work has been done. This project would involve investigating the properties of directed detachments.

Supervisor: Dr Keith Edwards


Artificial Intelligence

Argumentation in mathematics

This project will investigate the extent to which ideas in argumentation theory apply to mathematical proofs, in their construction, evaluation and acceptance. It is expected that such research will contribute to both argumentation theory by extending it to a non-standard domain and to the study of informal mathematical practice.

Supervisor: Dr Alison Pease


The automation of serendipitous discoveries

Serendipitous discoveries involves both chance and skill on the part of the discoverer to recognise the value of a new idea and to develop it into a useful discovery. This project will investigate whether it is possible to automate the process, and produce a computational model if so. The work will be firmly embedded within the field of Computational Creativity.

Supervisor: Dr Alison Pease


Modelling in Constraint Programming

Problems often consist of choices. Making an optimal choice which is compatible with all other choices made is difficult. Constraint programming (CP) is the branch of Artificial Intelligence, where computers help us to make these choices. Constraint programming is a multidisciplinary technology combining computer science, operational research and mathematics. Constraints arise in design & configuration, planning & scheduling, diagnosis & testing, and in many other contexts. CP can solve problems in telecommunication, e-commerce, electron- ics, bioinformatics, transportation, network management, supply chain management, and many other fields.

A constraint program consists of a set of variables, a set of possible values, for each variable and a set of constraints. For example, the problem might be to fit components (values) to circuit boards (variables), subject to the constraint that no two components can be overlapping. A solution to a CP is an allocation of values to variables such that none of the constraints are violated. A constraint solver searches for this solution by alternating phases of branching and inference. Typically the branching phase selects a variable and a possible value for it and seeks to find a solution in which the variable has that value. If no solution is found, then another value is tried. Branching thus causes the system to explore a tree of partial assignments, seeking one that can be completed. In the inference phase, the solver attempts to deduce consequences of the constraints and the current partial assignment and uses these to reduce the search. Modelling is the process of changing a problem from a real-world description to the variables, values and constraints that construct a CP.

Projects are available in the area of improving the modelling and solving processes in constraint programming.

Supervisor: Dr Karen Petrie


Space

Generating and enhancing artificial terrain in real-time on Graphics Processing Units

The proposed topic is to research and extend the limits of generating and enhancing artificial terrain in real-time using tessellation shaders which run on Graphics Processing Units. Terrain models can be obtained from incomplete or low resolution data sets and there is often a need to either increase the resolution or fill gaps with equivalent terrain. There are a variety of terrain interpolation and generation algorithms that can do these tasks but too slowly to be considered near real-time for excessively large data sets. Tessellation shaders provide a standard, accessible way to utilize the power of modern graphics hardware to modify polygon meshes and generate new polygons at far faster rates than could be achieved using fixed pipeline models. The aim of this research is to implement the current state-of-the-art terrain generation algorithms in tessellation shaders, characterize their performance and develop new algorithms to take advantage of the capabilities of graphics hardware. This research could be further generalized to apply to partial data sets and the techniques used to combine interpolation and noise to fill gaps with data in real-time.

Supervisor: Dr Iain Martin


Computer Vision

Activity Recognition using Computer Vision and Accelerometers

Algorithms will be developed and evaluated for automated recognition of complex manipulative activities using computer vision and accelerometers. This project will focus on food preparation as an example domain. Recognition of food preparation activities is challenging because it involves a variety of different objects, some of which undergo significant changes in form and appearance when manipulated. See http://cvip.computing.dundee.ac.uk/recog_food.ph... for further details of our group’s recent work on this topic.

Supervisor: Professor Stephen McKenna


Automatic Analysis of Microscopy Images in the Life Sciences

This project would develop algorithms for automatic detection, segmentation and recognition of subcellular organelles in microscopy images. The images are acquired using either electron microscopy or very high-resolution fluorescence microscopy. They reveal structures inside individual cells in great detail. The huge amounts of microscopy image data being generated by life sciences researchers require new methods for automated image analysis so that information relevant to biological and biomedical research can be extracted.

Supervisor: Professor Stephen McKenna


Automatic analysis of retinal images for clinical and biomarker pur-poses within VAMPIRE

VAMPIRE is an international project led by the universities of Dundee and Edinburgh. Its aim is to develop software to analyze the vasculature in retinal images (fundus, SLO) and to identify lesions relevant for high-incidence diseases. Various image analysis projects are available within VAMPIRE.

Supervisor: Professor Emanuele Trucco


Compressed Sensing for Video based Human Action Recognition

Human action recognition is a very important task in the area of visual surveillance, video search, health care, etc. However, despite the great success of text search, the video search is still in its infant. This there is a great incentive for the development of action recognition methods to boost those applications. One of the key tasks is to develop robust action descriptors and action learning methods. On the other hand compressed sensing has achieved great attention by the com-puter vision community. This project will take the recent research achievement on compressed sens-ing as a start and develop further novel methods for action recognition.

Supervisor: Dr Jianguo Zhang


Finding object of interest from a large image collections

As one of the fundamental tasks of computer vision, object recognition is becoming increasingly important. The research outcomes could provide security solutions in the area of visual surveillance and web based image retrieval. In past decades, numerous approaches have been pro-posed. Among them, bag of local features and biological motivated object models have become in-creasingly attractive towards solutions for real applications. The goal of this project is to develop more effective object recognition approaches by investigating several state of the art object models, particular on how to find effectively find object of interest from a large scale image collection that gathered by relevant search engine such as Google.

Supervisor: Dr Jianguo Zhang


Can't find what you're looking for? Contact Us