An application of the supervoxel-based Fuzzy C-Means with a GPU support to segmentation of volumetric brain images
In this paper we extend our previous work for planar images by adding a new step in the volumetric segmentation algorithm that allows us to determine regions closer to it. There are huge of papers for planar images and segmentation methods and most of them are graph-based for planar images and very few papers for volumetric segmentation methods. However, even if image segmentation is a heavily researched field, extending the algorithms to spatial has been proven not to be an easy task. A true volumetric segmentation remains a difficult problem to tackle due to the complex nature of the topology of spatial objects, the huge amount of data to be processed and the complexity of the algorithms that scale with the new added dimension. The problem of partitioning images into homogenous regions or semantic entities is a basic problem for identifying relevant objects. Visual segmentation is related to some semantic concepts because certain parts of a scene are pre-attentively distinctive and have a greater significance than other parts. A number of approaches to segmentation are based on finding compact regions in some feature space. A recent technique using feature space regions first transforms the data by smoothing it in a way that preserves boundaries between regions. The key to the whole own algorithm of volumetric segmentation is the honeycomb cells. The pre-processing module is used mainly to blur the initial RGB spatial image in order to reduce the image store and to make algorithms to be efficient. Then the volumetric segmentation module creates virtual cells of prisms with tree-hexagonal structure defined on the set of the image voxels of the input spatial image and a volumetric grid graph having tree-hexagons as cells of vertices. Early graph-based methods use fixed thresholds and local measures in finding a volumetric segmentation.