Fast and effective characterization of 3D region data

We consider the problem of characterization of spatial region data such as the regions of interest (ROIs) in medical images. We propose a method that efficiently extracts a k-dimensional feature vector using concentric spheres in 3D (or circles in 2D) radiating out of the ROI's center of mass. The proposed method can be applied to classification and similarity searches of ROIs. We also propose a region data growth model that we use to generate artificial data with various properties including homogeneous and non-homogeneous region data. We use the artificial data to evaluate the effectiveness of the characterization method comparing also its classification performance to mathematical morphology. The experiments show that the performance of our method is comparable or better than that of mathematical morphology although it is two orders of magnitude faster which makes it very suitable for application in very large image databases.