Erscheint 12 Ausgaben pro Jahr
ISSN Druckformat: 0040-2508
ISSN Online: 1943-6009
Indexed in
BUILDING DETECTION USING PROCESSING OF MONOCHROMATIC EARTH OBSERVATION IMAGE
ABSTRAKT
The paper presents a new technique for stage-by-stage image processing in order to detect buildings on monochromatic images obtained as the result of the remote Earth sensing in
various wavelength bands.
The method is based on the particularities of most man-made objects with the straight-line edges and positioned at right angles with respect of each other. Representing an image in the form of separately taken arrays of the size and direction of the brightness gradient allows detecting targets both in the cases with white objects at the black background and with the black
ones at the white background. The developed 2D adaptive image filter allows detecting straight edges of the objects even if the image fragment in question has a low contrast and a high level of noise. Further processing of the list of segments only, unlike processing of the entire bitmap
image, allows a substantial decreasing of the processing time.
At processing of the initial image, the developed algorithm forms up a limited set of objects, the shape of which is close to the man-made ones.
The suggested method can be used as an image segmentation algorithm as the basis for formation of hypotheses to provide for subsequent recognition using the neural networks. The method is also efficient if used as a means for automating and improving the efficiency of
preparation by the operator of the training image dataset for neural networks, as well as for optimization of the process concerned.