Research

Overview

Window in town hall,
Victoria Island, CA.
Shows Science, Medicine and Mathematics brought together.

My research area is image understanding and computer vision, especially in application to medical images. This research is interdisciplinary and it brings together studies of:

Physics of image formation

(How a structure in the body is transformed into an image? How well do these transformations preserve important features of the structure? How to recover information about the body structures from their images?).

Human visual perception

(How our visual system groups low level features into coherent entities? What human vision is good at and what are difficult visual tasks? How to engineer practical vision systems inspired by powerful mechanisms developed in natural vision?)

Medical diagnosis

(What visual clues clinicians use for diagnosis? How do they combine visual information in the image with previous experience to arrive at diagnosis? How to model and exploit these abilities to develop practical computer-based aids to diagnosis?)

In addition to interesting theoretical questions, the practical aim is to develop computer tools which assist the clinical expert in difficult visual tasks. Medical imaging domains include dermatology (pigmented skin lesions, or moles), ophthalmology (fundus images), endoscopy, radiology (X-ray mammography, CT and MRI images of the brain) and fluorescence and optical microscopy.

In collaboration with colleagues working in the area of Natural Computation I have begun to explore the use of genetic algorithms and evolutionary computation to evolve new methods and operators for image analysis and interpretation.

I am also interested in software engineering for computer vision. Work in this area includes requirement specification, validation methods and performance certification for computer vision software. Within this line of work we have developed methods for generating complex simulated images with strictly defined and controllable visual features which can be used either in computational experiments or with human subjects.