As a PhD candidate in computer science at Naturalis and LIACS, my research focuses on the development of computer vision algorithms for real-world biological challenges. The ultimate goal would be to create a computer vision system with an "eye for nature", transforming the way automated methods analyze the natural world.
Keywords
Computer vision, Deep learning, Mechanistic models, Bioinformatics, Biodiversity monitoring
Researchinterest
I am currently working on the TATTOOS project, which is an acronym for paTtern formATion Through biO-inspired cOmputer viSion. In this project, we combine biology and machine learning by integrating prior knowledge on biological pattern formation from mechanistic models into deep learning networks. We aim to create computer-vision algorithms that leverage the process of skin pattern formation for biologically informed predictions.
In recent years the collection of biodiversity data has exploded. Camera trap networks capture millions of wildlife images, drones map kilometers of animal habitats in mere minutes, and citizen scientist log thousands of observations on a daily basis.
Although this increase in data collection is more than welcome and much needed, the mere volume of data bars it from being manually analyzed. Computer vision provides a way to mitigate this issue. Algorithms can be used to e.g. separate objects from background, track objects of interest through multiple frames, or identify the contents of a frame.
This identification of image contents, also called image classification, performs quite well on the species level (for example, recognizing a cat as being a Felis Domesticus), but does not perform well on within-species individual identification (for example, recognizing a cat as being my cat Crunchy).
Animals like leopards, jaguars, ladybugs, house cats, giraffes, zebras, and zebrafish all have distinctive individual patterns. Developmental biologists have spent decades studying how these patterns for naturally as cells interact. In the TATTOOS project, we are taking that rich biological knowledge and baking it directly into computer vision algorithms. By teaching computers the biological "rules" of how nature designs these patterns, we hope to successfully identify individual animals in the wild and revolutionize how we protect biodiversity.
Keypublications
- van Bijsterveld, J., Avitabile, D., Verbeek, F. J., & Pucci, R. (2026). Exploring Clustering Capability of Inpainting Model Embeddings for Pattern-based Individual Identification. arXiv preprint arXiv:2605.04904.
PhDsupervisors
Related
