Automatic Extraction of Roads from Spot Images

摘要

The development of a system for semi automatic extraction of roads from SPOTimages is described. Digital image processing techniques supply automatic methods to process the huge amount of data acquired by satellites. Characteristics of the SPOT satellite, like a finer ground resolution, make research into the extraction of roads in satellite images more attractive. The idea is to create initially an interactive mode. Human experience obtained by controlling the algorithm is formulated in rules to automize more parts of the system. An overalll conclusion for previous work is that a combination of techniques is required for road extraction. Various techniques are organized in two levels. The two level contains three methods for segmentation: classification with multispectral and structural information, dynamic programming in a region of interest and profile analysis. The high level controls the segmentation level. This road interpretation level has four tasks: test the quality of results from the segmentation level; closer analysis of the local situation; recognition of road features; control over the segmentation level including start, stop, adaptation of parameters, sequence and integration of segmentation methods. The rules in the high level initiate a knowledge based approach. Techniques could be added to both levels. The type of landscapes in the SPOT data set, the urban agglomeration of western Holland, complicates the extraction of roads. The narrow grey level range of SPOT and the presence of texture, noise and striping influence the results. An effective preprocessing is desirable.