ADAPT Maynooth University researcher Jeremy Chopin recently published his PhD work within the journal Computer Vision and Image Understanding (CVIU) http://doi.org/10.1016/j.cviu.2023.103744 . Titled “Model-based inexact graph matching on top of DNNs for semantic scene understanding”, the work aims to improve 2D and 3D image segmentation. Co-authors include Rozenn Dahyot (Maynooth University), Harold Mouchère (Nantes Université), Jean-Baptiste Fasquel (Université d’Angers), Isabelle Bloch (Sorbonne Université).
This work proposes to postprocess semantic segmentation results (e.g. obtained by deep learning) with a graph matching approach for correcting potential over-segmentation errors. This post processing approach is very useful in particular situations where training data is scarce which can ultimately impact the accuracy of deep learning approaches. This novel approach is validated for 2D image segmentation and also for 3D volumetric image (MRI) segmentation.
The preprint paper is also open access on ArXiv here and readable in html here.
The central focus of the Computer Vision and Image Understanding Journal is the computer analysis of pictorial information. Computer Vision and Image Understanding publishes papers covering all aspects of image analysis from the low-level, iconic processes of early vision to the high-level, symbolic processes of recognition and interpretation.