The objective of this thesis is to go beyond the use of “basic” traces on each image to be segmented or textual descriptions that are not always usable by exploring methods aimed at guiding segmentation based on advanced characteristics (geometric, colorimetric, dynamic, etc.). The aim is therefore to propose a method capable of segmenting batches of images containing objects of similar interest (e.g., images of particles) without having to interact with each one, based on high-level features.