Voronoi Image Segmentation and Its Applications to Geoinformatics

Abstract

As various geospatial images are available for analysis, there is a strong need for an intelligent geospatial image processing method. Segmenting and districting digital images is a core process and is of great importance in many geo-related applications. We propose a flexible image segmentation framework based on generalized Voronoi diagrams through Euclidean distance transforms. We introduce a three-scan algorithm that segments images in O(N) time when N is the number of pixels. The algorithm is capable of handling generators of complex types (point, line and area), Minkowski metrics and different weights. This paper also provides applications of the proposed method in various geoinformation datasets. Illustrated examples demonstrate the usefulness and robustness of our proposed method.