DSVP Group: Research

Research

Vision

Computer Vision

The group's fundamental research in computer vision includes unsupervised segmentation, motion analysis, tracking, shape description, multi-scale medial axis transformation and model-based interpretation. The group is pioneering research in active colour vision with the goal of developing active measurement systems that do not need precision articulated heads to manipulate stereoscopic cameras.
[Brain Scan]

Medical Image Interpretation

This research is concerned with the analysis of biological cells, radiological images and efficient high-quality image transmission.

Industrial Inspection

This research is normally conducted in close collaboration with industry. Topics include petrochemical surveying, highway inspection and electronic assembly inspection. There is a strong emphasis on 3-D and multi-modal imaging techniques. Whilst it is often argued that the regular structure of man-made objects makes them less variable, in practice robustness can only be achieved using sophisticated imaging and computer vision techniques.

Speech and Language

Basic Research in Speech and Language Processing

Almost all current successful automatic speech recognition systems are based on the same types of statistical models, known as hidden Markov models (HMMs). Although HMMs have been extremely successful in terms of performance, they make significant unreasonable assumptions about the nature of speech. Our research in this area includes the development of new theory and computational models, which try to unify the different components of speech and language technology and overcome some of the limitations of current methods. Some recent examples include as Trajectory-based Segmental Statistical Acoustic Models and Products of Experts.

Application Areas

In addition to this fundamental research, we conduct research into specific topics in speech and language technology. As well as automatic speech recognition, these include speaker and language identification, speech and language technology for education (SLaTE), speech technologies for children, and accented speech (specifically speech which represents the regional accents of British English). We have collected the largest corpus of accented British English that is intended specifically for speech technology research.

Speech in Multimodal Interaction

Our most recent research is concerned with the integration of speech with other modalities, namely gaze and gesture. Our approach has been to apply pattern processing techniques from speech technology to these other modalities, and to combine them with speech in order to understand better an individual's intent. Much of this work has been done in collaboration with the School of Psychology at the University of Birmingham, using their expertise and systems for measuring eye-movements and 3D body motion.

Hardware

Novel Architectures

This research addresses novel architectures for coarse and fine-grain parallelism. Research on coarse grain parallelism has resulted in the development of a novel heterogeneous architecture which integrates numeric and symbolic processing, while the focus of our research in fine-grain parallelism is reconfigurable systems. This research is attracting enthusiastic support from major companies in the field.

Acoustics

This is a long established group with an international reputation, working on all aspects of underwater acoustics and sonar. More information on current research in acoustics is available on the Acoustics and Sonar page.

Electronic, Electrical and Computer Enginnering