The Geographic National Institute (GNI) map in the scale of 1:25,000 is made out of four color plates. The green one represents the vegetation, the blue one stands for hydrography, the black is for road and toponymy, and the orange color for other various information. Our work is based on the scanned green plate. The purpose of our study is to segment and extract different zones of vegetation so as to integrate them in a GIS (geographical information system). The image is divided into untextured and textured solid areas. On this kind of image, two categories of textured regions can be found: macrotexture, represented by a set of similar graphic elements and microtexture which is represented by a set of small points or thin lines located very close to each other. There is also a superposition of micro and macro texture. These textures are synthetic, regular and often periodic. The major issue is to find a general method which can be applied to different representations. Our segmentation method is driven by the dichotomy decision. Cartographic redaction implies some visual perception rules. To be informative, variations on maps are limited: intensity, coarseness, orientation. We work on a hierarchical classification of the different areas: textured or nontextured regions, coarse or fine textures as well as intensity, orientation and periodicity of textures. Operators are associated to this hierarchy. We use statistical approach for microtextures and structural approach for macrotextures. Different texture analysis methods are tested and some have to be modified to be adapted to this kind of image. In fact, even if the image has 256 gray levels, only few gray levels are informative. For graphic elements, we focus on geometry properties. The other problem is to detect boundaries among areas. The limits are drawn sometimes but often vegetation regions are bordered by roads or rivers which are present on the other color plates. That is why, in case of the subjective boundaries, we need information from other plates to confirm our results. We noticed that for fine textures, the edge could be relieved. The first experimental tests have demonstrated that our procedure is efficient.
|