Machine learning guided image reconstruction: A hybrid, classification - inversion approach for tomography

The Computational Imaging and Data Analytics Group of the School of Engineering at the University of Edinburgh is looking for motivated and talented students who are interested in pursuing full-time research towards a PhD in a new and exciting area of inverse problems at the interface of image reconstruction and analysis. This project will focus on limited data absorption tomography that finds numerous applications in biomedical imaging, environmental monitoring and non-destructive testing.

Conventionally, image analysis in the form of pattern recognition, segmentation, feature extraction, or classification follows image reconstruction and it is aimed at enhancing the quality and resolution of the image in order to allow for effective diagnosis and interpretation. In this sequence of events, one first addresses the reconstruction problem to form the image, typically by solving a model parameter estimation problem for the available tomographic data. Thereafter in the analysis stage the resulted image is processed independently without considering the tomographic data or the reconstruction algorithm it has originated from. This project will investigate way under which aspects from the reconstruction and analysis stages can be reconciled in order to develop new paradigms that exploit the fidelity of machine learning methods and the efficiency of data fitting inversion algorithms. Emphasis will be given to the case where the tomographic data set is limited or corrupted, and therefore to image at sufficient spatial resolution necessitates introducing prior information on the target’s composition and expected dominant features which will be imparted using clustering and segmentation schemes. At the same time, these machine learning tools will be adapted to ensure that clustering and segmentation is consistent with the tomographic data.

The implementation of the project requires algorithmic design, coding (in Matlab or Python), mathematical analysis to verify the performance of the algorithms, as well as dissemination of the methods in conference and journal publications. The successful candidate will join a vibrant community of researchers working in signal and image processing and will benefit from the interaction with our academic and industrial collaborators. There will also be opportunities to travel to conferences and attend workshops relevant to imaging sciences, inverse problems, and data science.

Further Information: 

Closing Date: 

Thursday, May 10, 2018
Engineering School Logo
Engineering School Logo

Principal Supervisor: 

Eligibility: 

Minimum entry qualification - a first class degree (or International equivalent) in engineering, computer science or mathematics and/or a Masters degree in a computational discipline. Further information on English language requirements for EU/Overseas applicants.

Funding: 

Tuition fees and stipend are available for Home/EU students (International students can apply, but the funding only covers the Home/EU fee rate).

Further information and other funding options.

Informal Enquiries: