Researcher: Rihuan Ke, Carola-Bibiane Schönlieb and Peter Schuetz
Microscopy images are essential tools in the study and analysis of the food microstructures. While high-quality images of food can be efficiently acquired by modern microscopy techniques, significant image processing (e.g., segmentation) is needed to obtain reliable quantitative information for the microstructure. However, this is usually not an easy job, because many food materials, especially soft solid, have complex structures with multiple phases that are hard to differentiate. It makes many currently available software or toolbox fail to work at a satisfying level. In this joint project with Unilever, we are interested in analysing SEM (scanning electron microscopy) images of ice cream. Ice cream is made up of ice, air, fat, and a concentrated unfrozen solution. The quality of ice cream is determined by the proportion of these components as well as the arrangement of ice crystals and air bubbles. The detection of the elements and their interface is therefore a key factor for the innovation of ice cream manufacture.
In this project, we are developing machine learning frameworks, in particular deep learning methods, for automated classification and segmentation of the ice cream images.