Road Extraction from Color Imagery

The detection of boundaries in the field of digital photogrammetry and image analysis is still a challenging area of research. Contour models recently gained a lot of attention for recovering and localizing edge structures and thus are useful for facilitating the image segmentation problem. Their use has been extended for describing man-made structures such as roads and buildings from gray-level imagery. In this paper, the problem of incorporating color image information mainly of R,G,B content in deformable models of contour-type is addressed for the extraction and localization of road structures of small width and sequentially for supporting the verification and revision stages of the mapping process. More precisely, two models will be discussed together with their mathematical modeling and their implementational issues. In the first model, the image energy term is formulated as a function of regularized-type by transforming the original R,G,B image content to color intensity and contrast energy terms which finally, contribute unevenly to the radiometric content. The second model minimizes the given initial description of the road structure as the energy flow on the color gradient field coming from the fusion of the R,G,B content while preserving color invariance. The solution is taken as the optimum 'color weighted path' after reaching global minimum for the whole content of the road structure and is iteratively executed. In both models the solution presents the final road centerline shifted and repositioned which is localized in sub-pixel means taking into account peculiarities and characteristics of certain spatial extent. Experimentation with medium scale and SPOT satellite imagery, presenting roads in semi-urban and in forested areas, show the performance and evaluation of the above models.