Source: Calissano, Anna, Aasa Feragen, and Simone Vantini. "Populations of unlabeled networks: Graph space geometry and geodesic principal components." MOX Report (2020).

Helping make better pizza


We can probably all agree that pizza is one of the most universally tasty foods, especially when the cheese melts just the right way. But what is a good Mozzarella cheese? There are many answers to that question, but it turns out that up till now nobody looked deep enough, to get a full understanding of why one cheese is better than another.

This PhD is a part of a bigger project that aims to improve our capabilities in analyzing the microstructure of Mozzarella cheese and correlating them with its functional properties. All that will let us optimize the production process, not only to produce better and more consistent cheese, but also to limit its environmental impact.

We expect that our work will provide a direct description of the cheese microstructure, which can then aid the decision process during production. We plan to acquire a diverse dataset of µCT (Micro-Computed Tomography) images of mozzarella cheese samples. Based on the assumption that mozzarella cheese is anisotropic, we expect to observe a distinct microstructure of fats and proteins that can be extracted and correlated with functional properties of each sample.

The first challenge of the project is to successfully acquire the cheese-image dataset. In order to avoid the deterioration of the cheese, we need to ensure a cool and stable environment during prolonged CT-scanning periods (even up to 20 hours per sample). Additionally, the attenuation coefficient of fats and proteins is very similar, so the scanning parameters have to be carefully adjusted to best visualize the transition between them.

Since we aim for explainability, the first attempts to extract the structural information will be based on a structure tensor analysis. It outputs a local orientation of fibers in a material, which can then be compared across samples. This will form the baseline, before moving to deep learning-based methods for extracting quantitative information from volumetric data.

The methods and algorithms developed in this project should be generalizable to other materials apart from cheese. Since fibrous materials are widely used in e.g. insulation, packaging, textiles or composites, this work will have a large impact on the topic of quantification of structural properties and their use in inferring functional properties.

PhD project

By: Paweł Pięta

Section: Visual Computing

Principal supervisor: Anders Nymark Christensen

Co-supervisor: Anders Bjorholm Dahl

Project title: Extraction and analysis of Mozzarella cheese microstructure based on µCT images

Term: 08/2022


Pawel Tomasz Pieta
PhD student
DTU Compute


Anders Nymark Christensen
Associate Professor
DTU Compute
+45 45 25 52 58


Anders Bjorholm Dahl
Professor, Head of Section
DTU Compute
+45 45 25 39 07