G-SNAKES: A NEW ALGORITHM APPLIED TO OIL-SPOT SEGMENTATION USING RADARSAT 1 IMAGES
C.M.F. Pereira1, E.C. Riqueza2, F.L. Melo3, O.R. Vergara1
1 - IME, Cartographic Engineer Department, Rio de Janeiro,
Brazil
2 - IME, Chemistry Engineer Department, Rio de Janeiro,
Brazil
3 - CTEx, Tecnologic Center of Army, Rio de Janeiro, Brazil
cassius27@gmail.com
Accidental pollution causes the
launching of a great polluting mass in the environment, resulting on
considerable damages. Among many types of accidental pollution, those involving
oil spills requires a great concern because of its fast dispersion. Although
difficult to be evaluate, the definition of the accident localization and
magnitude becomes a very important. Due to this complexity, this paper proposes a methodology for using measures of texture and
deformable models in order to identify and quantify oil-spot areas. A new
image segmentation algorithm named G-Snakes was developed, based on. deformable
models, known as geodesic active contours, that consist in deforming an initial
curve to identify and extract the contours of objects of interest in images.
The proposed algorithm intends to minimize inconsistent results, such as
over-segmentation of homogeneous areas, generation of false boundaries and open
contours. These incoherent results are usually associated to traditional
methods of segmentation such as clustering and region growing. Another
advantage of the implemented algorithm is its outputs of cartographic features
boundaries in vector format, which makes possible any further necessary edition
of these data. Considering the area features, the implemented
algorithm produces not only the feature contour, but also computes its surface.
As a study case, this paper analyzes an oil spill in Guanabara Bay, Rio de
Janeiro, Brazil occurred in January 18th, 2000. The methodology for
the oil-spot area extraction consists on three tasks: 1) Processing of a SAR
image with textures filters, for speckle attenuation; 2) Applying the G-Snakes
algorithm in the SAR processed image for the oil-spot segmentation; 3) oil-spot
area calculation. This methodology proved to be efficient for performing a
quick evaluation of the disaster, and for supplying information about
localization, shape and dimension of the oil-spot. These parameters are
essential for monitoring any kind of environmental disaster.