We’re developing novel theories and algorithms to capture, process, analyse and understand single and multiple images and videos. Our research covers a diverse range of areas including:
- analysis of remote sensing imagery to shape our knowledge of the impacts of climate change
- medical imaging to help shape the design of future artificial knees
- video and image compression
- real-time imaging
- adaptive optics
Our research has been instrumental in the development of a new multi-modal similarity measure called the Sum of Conditional Variance (SCV). The SCV is the first information-theoretic similarity measure that can be optimised using standard optimisation techniques. The SCV similarity measure is adopted for use by leading international research teams including:
- Johns Hopkins University (USA), where it is being used for visual tracking of robotic surgical tools in retinal surgery
- University College London and Imperial College London who are using the SCV similarity measure in a new technique for real-time tracking of deformable tissue surfaces during keyhole surgery
- a team in France that developed a modified version of the SCV for the real-time tracking of soft tissue in 3D ultrasound sequences. The team members hold positions at the Institut de Recherche Technologique, Inria, INSA, CHU, Université de Rennes and IRISA.
Our research on satellite image analysis has a strong impact on converting remote sensing raw data into usable information across two aspects:
- the advanced investigation of image quality improvement via postprocessing, including error corrections such as multisource image registration and optical data cloud removal, super-resolution reconstruction and subpixel and super pixel-based techniques
- the development of effective machine learning as well as deep learning approaches. Reliable training of a classifier, which is critical for accurate information extraction, is addressed by the development of feature mining techniques in both spectral and spatial domains and the generation of novel class data models in high dimensional space.
We have world-leading expertise in:
- 2D and 3D medical image registration for kinematic analysis of knee and hip joints
- real-time imaging for autonomous guidance of unmanned aerial vehicles
- feature extraction and classification of hyperspectral imagery
- motion estimation for video and image compression
- algebraic reconstruction techniques for faster and finer CT reconstruction
- adaptive optics for turbulence characterisation and retinal imaging.
Our collaboration with the Trauma and Orthopaedic Research Unit (TORU) at the Canberra Hospital since 2005 has developed more effective methods to measure the relative motion of the bones in human joints using standard hospital imaging equipment. The major outcome of this project is the development of a software package called OrthoVis.
The algorithms used in OrthoVis are a result of several PhD student projects supervised by members of the Imaging group. This software is now an intrinsic part of a major clinical trial to evaluate the performance of three different artificial knee designs.
We also demonstrate the applications of remote sensing imagery across:
- crop mapping
- weed classification
- flood assessment
- moving vehicle detection from satellite videos.