Object detection in satellite images, vectorisation of multispectral satellite images with mathematical morphology
Satellite sensors naturally produce raster images. However, a vectorised image has certain advantages, most notably the ability to be scaled and transformed with no loss of information. Algorithms are required to convert satellite raster images to a vectorised form. Existing software packages either cannot export to Scale Vector Graphics (SVG) or they are not able to store the geographical information within the SVG. Therefore, a new software product is needed. This paper introduces vectorisation algorithm for multispectral satellite images using mathematical morphology. It starts with a segmentation method, which processes all channels of the input satellite image in parallel. The boundaries of the objects, detected by the segmentation step, are extracted afterwards. These boundary lines can optionally be simplified for data reduction and smoothing purposes and represented as polygons. These data, together with attributes from the objects of the segmented image and the georeferencing and projection information, are stored in the vector file. It is shown that the result of the algorithm is similar to that of expensive commercial software products, while it supports ESRI shapefiles as well as SVG in which also the georeferencing information are stored.
Bibliographic Reference: EUR 21293 EN (2004), 119 pp. Free of charge
Availability: Available from European Commission, JRC Knowledge Management Unit, Ispra (IT) Tel: +39 033278 9893 Fax: +39 033278 5409 E-mail: firstname.lastname@example.org or Bruxelles (BE) Tel: +32 2 295 76 24 Fax: +32 2 299 63 22
Record Number: 200417801 / Last updated on: 2004-11-18
Original language: en
Available languages: en