Analyse d’images 3D

Developing appropriate 3D image analysis tools is often necessary to extract relevant data from snow tomographic images.
Original computation methods (normals, curvatures, surface area...) are presented below.

 Normal vectors

Applicable on binary images, the presented algorithm optimizes the smoothing of digitization effects while preserving relevant details of the processed numerical object. It gives an accurate normal field, which can be used for all kind of 3D numerical measurements.

- Main related reference :

 Surface area estimation

To obtain the surface area, the simplest method consists in counting the surface voxels of the 3D object. This method is inaccurate because the physical surface that is represented by a voxel depends on its orientation along the coordinate axes. In 3D, this can induce an error on the surface estimation about 200%. Our method takes into account the surface orientation in each voxel and gives much more precise surface area values. To obtain accurate surface estimations, a precise determination of the normal vector in each surface voxel is required.

- Main related reference :

 Mean curvature

The mean curvature is a fundamental parameter for snow metamorphisms. With our method, we can obtain the 3D curvature on each voxel of the object’s surface.

- Main related reference :

 Gaussian curvature

The gaussian curvature is a key-parameter to describe the geometry of snow. In particular, it allows to detect neck regions between grains. With our method, we can obtain the 3D Gaussian curvature on each voxel of the object’s surface.

- Main related references :

 Grain segmentation

Deposited snow on the ground consists of ice grains that have been welded by thermo-dynamical and mechanical constraints. To analyse the snow microstructure and to understand its formation and its time evolution, it is often required to numerically separate these ice grains from their neighbours. Commonly used three-dimensional imaging methods do not provide any direct information on the boundaries between grains. To obtain an accurate segmentation, we developed several geometrical methods.

- Main related references :

 Minimum cut

The minimum cut surface that disconnects two opposite faces of a snow sample allows for a good characterization of its bonding system. We implemented a graph-based algorithm to compute such a surface for any 3D image in the x, y and z directions.

- Main related reference :

Voir en ligne : Microstructure de la Neige