IPI - Image Perception Interaction
Discreet representation of information
This theme is centered around discrete geometry.
We are particularly interested in discrete tomography with the Mojette transform and similar discrete Radon transforms such as the FRT (Finite Radon Transform) as well as discrete distances and mathematical morphology.
These subjects have applications in medical imaging and non-destructive testing.
We wish to reinforce the international visibility of the team, as demonstrated by the organization in 2016 of the international conference Discrete Geometry for Computer Imagery, a reference in the field.
We are continuing work on the Mojette transform and the discrete Radon transforms. For the aspects of using the redundancy of the Mojette transform, we will explore the theoretical furrow of the nature of zero space and ghosts, especially with our Australian colleagues. This will be amplified by the study of the 3D transform and its use for exact rotations.
We also wish to develop the study of these transforms on the networks of regular non-cubic points. A more efficient representation of information is expected due to the use of denser regular grids (especially for dimensions greater than 3). The links between the different discrete transformations of Radon, between them and with other transformations, will be deepened. We expect both theoretical developments and algorithmic progress.
The development of fast discrete distance calculation algorithms and mathematical morphology operators remains a crucial point in order to be able to respond to real-time applications and very large volumes of data. Moving from the binary case to the morphology in ternary (or even quaternary) is also already required by medical or vision applications.
The current and future results will be integrated into applications such as medical imaging with the many partnerships existing at this level in Nantes. For Mojette tomography, we need to validate existing results with visual quality metrics and combine it with other discrete geometry tools (such as rotations) into a coherent and reusable set. More broadly, we will also try to use the Mojette transform and discrete geometry to represent and manipulate media information in a different way.
Part of our work will focus on the analysis and segmentation of medical images. For example, we will be interested in diagnostic assistance through automatic detection of ischemic lesions in neurology and cardiology (collaboration with Pr. H. Desal and JM. Serfaty from Nord Laënnec Hospital). This work is part of L. Mahé's CIFRE thesis with Keosys. We will continue our work with Professor Y. Amouriq (in dentistry) on the segmentation and characterization of trabeculae (bone textures) and the vascular tree. This collaboration, initiated in 2010, has already led to 11 co-publications and several co-supervision (Thesis/Master). Finally, we will focus on the 3D characterization of muscle cells. Following confocal microscopic acquisitions on animal cells (dogs) at the Nantes Veterinary School (ONIRIS), confocal images (cells) will be segmented and classified (membrane, nucleus, proteins, blood vessels).
During this work, we may need to study image registration methods before calculating inter-image similarity measurements. In addition to traditional segmentation methods (such as active contours or mathematical morphology), we will seek to use perceptual quality evaluation metrics.