Geoscience Reference
In-Depth Information
scene image and a series of templates is computed through an implementation of
FFT, which makes this problem a popular question in this field, including image
processing and pattern recognition. The radii of the implemented templates vary
from interior minimum 5 pixels to the maximum 100 pixels. The correlation will be
easily obtained and be taken as the probability that a given pixel is the center of a
crater with radius r.
Crater Detection
The identification of impact craters on an image is achieved through the analysis
of the probability volume created as a result of the template-matching procedure
that was followed in the previous phase. Each element of the probability volume
is surrounded by six neighbors considering the 3D connectivity, and only the value
of above 30 % will be remained as potential candidates for the next step analysis.
The probability volume is used to compute the regional maxima by dilation. They
reflect the presence not only of interesting features that most like craters but also
some structures of other origins which results in some false detections. The maxima
can be computed using a two-step procedure.
1. H-maxima transform is applied to suppress all maxima whose heights are smaller
than a certain value h .
HMAX h .P/ D R P .P h/
(6.3)
2. The extended maxima EMAX can be given on the following after the first step:
EMAX h .P/ D RMAX Ĺ’HMAX h .P/
(6.4)
The signature of a crater can be found in the horizontal planes of the volume;
each corresponds to a given radius. Only those that contain extended maxima must
be analyzed. Two more steps will be undertaken in order to identify those that are
actually markers of craters. First, use an area-opening transform to filter the small
objects that cannot be craters. Second, analyze the roundness of the remaining
objects, and those objects whose roundness outside a given range (0.76-1.8) are
eliminated. At the end of this sequence, the point with coordinates ( u , v ), which
corresponds to the center of mass of each surviving object, is considered to be the
center of a crater with radius r and probability p .
6.2.1.2
A Modified Adaboosting Method
Adaboosting approach is a widely used method in the field of pattern recognition,
which is proved to be very effective for the automatic face detection (Viola and Jones
2004 ). Some results have been obtained in the field of crater detection based on this
Search WWH ::




Custom Search