Region growing in GIS, an application for landscape character assessment

Region growing is an image analysis technique that uses the spatial patterns in an image to divide an image into regions or spatially continuous clusters. A region growing process is an iterative bottom-up optimization process, where in each round the most similar neighbouring regions are merged, minimizing the dissimilarity within the regions and maximizing the dissimilarity between the regions. The underlying data structure of a region growing process is the regional adjacency graph (RAG). A RAG = (V,A,W) is a mathematical data format in which the regions are represented by a set nodes V = {v1, v2,...,vn}, the adjacency between these nodes by a matrix matrix A where akl e {0,1} and the dissimilarity between the nodes by a matrix of weights W, where wkl e R. In geo-sciences region growing is mainly used for the classification of high resolution satellite imagery and aerial photographs. By representing a spatial database as a RAG the scope of region growing can be broadened to other types of spatial data, including raster and vector data consisting of nominal and continuous data, operationalizing a new tool for the analysis of spatial data patterns in a GIS. This tool can be used for the classification of mosaic patterns on a vegetation map, to detect clusters in point clouds or to aggregate spatial data for map generalization and other applications where the segmentation of spatial data patterns is required. We have implemented a region growing algorithm in a methodology for landscape character assessment. We define the landscape character as presence, arrangement and variability of different landscape features. A region growing algorithm is used to delineate landscapes in a spatial database of landscape features. Each region represents a distinctive landscape and is characterized by its pattern of data values. The quantitative description of the landscape character can then be used to further analyse the dataset. The methodology proposed consists of 4 steps: 1. Building a spatial database 2. Delineating landscapes 3. Classifying landscape types 4. Evaluation of the analysis result.