A satellite image used to forecast forest fires carries no indication of how reliable its measurements actually are. This fellowship aims to fix that gap by advancing the mathematical and computational methods behind quantitative imaging—images that serve as measurements, not just pictures. The core problem is that raw data from medical scanners, satellites, and radars is always corrupted by noise and limited in resolution. Even the most sophisticated imaging methods today can produce accurate estimates, but they cannot reliably quantify the uncertainty in those estimates. Without error bars, decision-makers cannot judge how much trust to place in the image-derived evidence. If successful, this work would give doctors, climate scientists, disaster responders, and quality-control engineers a rigorous way to attach confidence levels to their imaging data. That would transform quantitative images from useful approximations into properly calibrated scientific evidence. The research is primarily fundamental—it builds the mathematical and computational foundations for future imaging technologies—but its impact would be felt wherever high-stakes decisions depend on what an image actually measures.
View original technical description
Digital images inform decisions that have a major impact on the economy, society, and the environment. Illustrative examples include decisions in medical practice, disaster recovery, agriculture, forestry, climate action, quality control, pollution monitoring, and defence. Such images are predominantly generated by using specialised devices (e.g., medical scanners, telescopes, and radars), which leverage decades of progress on sensor technology and instrumentation engineering. In addition to state-of-the-art hardware, modern imaging devices also rely strongly on sophisticated mathematical techniques and computational algorithms in order to transform the acquired data into high-quality images and extract useful information from them. Whereas most of us utilise photographic images as visual reminders or for social reasons, the images acquired for the aforementioned purposes are often used in a quantitative manner, as measurements of high-dimensional physical quantities of interest, so-called "quantitative imaging" (e.g., a satellite image measuring a vegetation index that is used to calculate water deficit for forest fire forecasting). Quantitative images are used as evidence in decision making or as scientific evidence, and are therefore subject to high accuracy and robustness requirements. This fellowship proposal focuses on advancing the mathematical and computational foundations that will underpin future quantitative imaging technologies. A major challenge facing quantitative imaging sciences is that the analysed data does not contain enough information to accurately determine the exact value of the images (the data is corrupted by measurement noise and has limited resolution). The last decades have witnessed remarkable progress in quantitative estimation accuracy. However, in addition to accurate solutions, decision-making and scientific research also require a precise characterisation of the uncertainty in the delivered solutions (e.g., the degree of certainty or "error bars" attached to the water stress predictions that underpin the potential for forest fires). Unfortunately, even the most sophisticated quantitative imaging methods currently available are unable to accurately quantify the uncertainty in their solutions. This critical limitation severely hinders the value of quantitative images as evidence for decision-making and science.
Plain English summaries and category classifications on this site are generated by AI and may not perfectly reflect the original research.
Is something wrong? Let us know