Automatic Recognition of Impact Craters on the Martian Surface from DEM and Images - Planetary Exploration and Science: Recent Results and Advances - page 108

6.2.2.2 Crater Detection by Using Morphological Characteristics

Detection of Crater Centers

Two different strategies are developed in this process to detect the centers of
simple and complex crater candidates according to their different morphological
characteristics (Xie et al. 2013 ). In the first strategy, we can identify and extract
the crater centers by local minima filter. Ordinarily, a pixel will be recognized
as a local minimum if its value remains the same after being filtered. Here, we
use a multi-scale square window to search the test image. However, during the
experiments, there are usually several connected minima located at the bottom of a
single crater. This will reduce the method's efficiency in the following steps. Thus,
during filtering, if another minimum has already been detected within the extent of
the filter, the new one will be marked and finally removed from the results. The
radius of the filter determines the minimal size of detectable craters. To examine the
method's ability to detect small craters, the radius of the filter is set to three pixels
in this letter as a crater's morphology cannot be effectively presented with fewer
pixels.
For the complex craters with central peaks or other structure instead of local
minima, it is necessary to develop another strategy to handle them. In this section,
we will detect the crater centers by using the features they have in common in the
slope matrices of DEM. No matter what structures a complex crater possesses, there
is always a flat and symmetrical area between its wall and center. Even in complex
situations, when craters are superposed and intersected, the flat areas of different
ones will be split by the walls and thus can be easily distinguished as well.

Detection of Crater Rims

After the detection of crater centers in the first step, we will detect the rims of simple
craters and complex craters in two different datasets, respectively. The original
DEMs are used as the basic data to detect the rims of simple craters. In the radial
direction of a detected crater center, a local maximum can represent the rim of the
crater. Thus, a complete revolution of scan lines from a crater center is used to detect
and mark the rim points. However, numerous pixels will be repeatedly scanned on a
frequent basis, and this will greatly reduce the efficiency of the CDA. Thus, a square
scan frame is used to replace scan lines and the length of the square will increase
gradually. Here the original side length of the frame is three pixels. The following
points in the same direction will be marked and no longer be checked after a rim
pixel is detected.
When it comes to complex craters, there are much more complex local variations
in DEMs, particularly the central peaks near the centers, disturbing the process of
detection of the radial local maxima. Thus, the slope matrices are used instead.