Automatic Recognition of Impact Craters on the Martian Surface from DEM and Images - Planetary Exploration and Science: Recent Results and Advances - page 100

binary image of the candidate area is submitted to a template-matching procedure.
Therefore, there will be a probability volume, 3D matrix constructed for further
analysis to determine the location and dimension of the impact craters that are
present in a test image.

Candidate Selection

Because of the crash of bolide on the planetary surface, the impact craters are
bowl-shaped depressions due to the propagation of a shock wave. As a result, they
show a roughly regular circular shape when imaged from orbit. Because the marked
contours by the elevated crater rims produce shadows, the difficulty in correctly
detecting craters has increased. Furthermore, the shadows are far from regular so
that they cannot be used as recognition features applied in detection. Therefore,
some preprocessing steps need to be done in this section.

1. A mask M with n n elements is centered over each pixel I uv of the original
image. The local mean m , the minimum and maximum of the values included in
the mask, will be computed, which produce a new image A defined by Eq. 6.1 .

A uv D max Œm.M/ min.M/; max .M/ m.M/ (6.1)

2. After the first step, a threshold value T can be obtained.

T D ǛŒmax .A/ min.A/ C min.A/ (6.2)

where Ǜ is a constant (0.25), which is chosen after a thorough search for the best
possible results for this type of image.

A binary image will be created when the threshold is applied to compute the new
pixel B uv and it is of highly contrasted local edges. This image is considered to be
the best candidate for the application on the following steps.

Matching

After the first step of preprocessing the image, it is necessary to locate impact
craters obtained in the previous phase. A simplified binary model of the specified
features is constructed to serve as a template for the matching of craters rims. It is
also necessary to incorporate a thickness which can represent the irregularity of the
natural shapes of impact craters. Therefore, a binary template applied in this section
is constructed like a white crown that is surrounded by a background in the shape
of a black square, and each radius must be analyzed to be a corresponding different
template.
Many matching methods can be applied for the purpose to calculate the similarity
of the template and the scene image. In this section, the correlation between the