A Multilevel Segmentation Method

Segmentation is an essential ingredient in a wide range of image processing tasks and a building block of many visualization environments. Many known segmentation techniques suffer from being computationally exhaustive and thus decreasing interactivity, especially when considering volume data sets. Multilevel methods have proved to be a powerful machinery to speed up applications which incorporate some hierarchical structure. So does segmentation when considered on quadtree respectively octree data sets. Here we present a new approach which combines a discrete and a continuous multilevel segmentation model. In figure 1 four different grid segments are depicted which resulted from the multilevel segmentation by using combinations of different boundary indicators which will be described later in detail. At first, the discrete method enables a fast segmentation depending on possibly multiple parameters describing the segment boundary and on selected seed points inside a segment. In an interactive process the user is able to ajust seed points which steer the automatic discrete segmentation process. Furthermore fast multilevel splatting techniques simultaneously enable interactive frame rates in the visualization to validate the obtained results. Thus, the user is effectively supported in the selection of appropriate parameters for the segmentation. Once an acceptable voxel discrete approximation is found a second segmentation and smoothing method based on a continuous model comes into play. It can be regarded Figure 1: Segmentation Results by greyvalue and gradient magnitude thresholding and the exclusion indicator σ = −χGc to prevent the algorihm to grow into Lena’s hat. Efficiency: 0.152. as an suitable postprocessing step. Hence, solving an appropriate diffusion problem the boundary approximation of the already obtained segment is improved including a suitable tangential smoothing.