Twin Snakes for Semi-automatic Line Extraction in Mid- and Large-scale Digital Photogrammetry

Line extraction is a basic tool required in various digital photogrammetric tasks. Image lines are elongated, narrow structures in digital imagery that can be clearly separated from the background. Typical examples of lines in aerial imagery are roads that usually appear as bright bands in a dark background. Besides roads, other linear features can be extracted such as railways, rivers, long, narrow fields, etc. Extracted data can be used for map production or for update of geographical information systems (GIS). In some close-range photogrammetric applications lines have to be extracted, too. In mid and large scales the lines have a significant width of some pixels, so that the two boundary curves can be extracted separately. Nevertheless, it is advantageous to make use of the parallelism constraint between the two curves during the extraction process in order to get a reliable result. For semi-automatic (user controlled) edge extraction in digital imagery snakes are widely used. Snakes are curves with an energy function assigned to them. They move in the image and change their shape, until they have found a minimum in their energy function. Then, the state of the snake (its position and shape) corresponds to an image edge. In this research project two such snakes, the twin snakes, that are coupled in their energy function, are used to detect the parallel boundary curves of lines. Furthermore, other sources of data will be integrated into the extraction process, e.g. the data extracted from other images.

In this research project, a new approach for finding optimal seam lines in orthophoto mosaicking by use of snakes has been studied. In the proposed algorithm, seam lines for adjacent orthophotos are defined fully automatically in overlapping areas of maximum similarity. Snakes are powerful tools for automation of line extraction from digital imagery. Computer learn to find lines such as roads or field boundaries. Twin snakes are two such snakes coupled for finding parallel curves. A snake is a model of the curve to be detected in the image. It continuously moves through the image while changing both its position and its shape until the curve has been found. This evolution of the snake is controlled by minimizing an energy function. Snakes can be used for various photogrammetric tasks. Some of them have been investigated in this research project. In particular, the theory of snakes was used for finding optimal seam lines in orthophoto mosaicking. Orthophotos (orthoimages) by definition are photos (in general aerial photos) from which geometric distortions (tilt and relief displacements) have been removed and which thus have geometric properties like maps (ground projection). For creating an orthophoto mosaic, neighboring and partly overlapping orthophotos of a scene are merged. This should be done in a way that the transition from one to another orthophoto can not be seen. The production line of orthophotos consists of several steps, each of which can introduce a different appearance regarding geometry, radiometry and spectral properties to the resulting orthophoto. For mosaicking adjacent orthophotos, a path of lowest difference in a combination of criteria is searched in the overlap area of these images. The seamline is chosen along this path of maximum similarity. In this research project, a strategy was developed to find this path of maximum similarity by use of snakes. The energy function of snakes was formulated in a way, that low energy represents areas of high similarity. On the other hand, high differences in overlapping regions of adjacent orthophotos (in intensity and hue of colors or in the texture of a scene) are penalized by high energy. This formulation of the energy function makes snakes detect the path of maximum similarity. Details can be read in the "International Journal of Photogrammetry and Remote Sensing", Vol. 56/1, pp. 53-64. (Kerschner M., 2001. Seamline detection in colour orthoimage mosaicking by use of twin snakes).