Automatic Recognition of Impact Craters on the Martian Surface from DEM and Images - Planetary Exploration and Science: Recent Results and Advances - page 99

In this chapter, the recognition methods of global Martian impact crater catalogue
are reviewed and presented in the following. At the same time, their performances
and shortcomings in detecting craters are discussed. A modified adaboosting method
and results of automatic recognition of impact craters on the Martian surface are
presented and discussed.

6.2 Recognition Methods

6.2.1 Recognition from Image

The exploration of the solar system by automated probes has obtained large numbers
of images of the surfaces on planets and satellites. Image analysis plays an important
role in archiving, retrieval, processing, and interpretation of large amounts of
image data as well as classification of all the resources. However, it is difficult to
automatically recognize impact craters precisely from the planetary images due to
the lack of distinguishing features, heterogeneous morphology in images, and huge
amount of sub-kilometer craters in high-resolution planetary images (Kim et al.
2005 ).
There are a number of techniques in the field of image processing and pattern
recognition with the purpose of the automated detection of impact craters from
images of planetary surfaces. Most previous studies focused on a given technique
primarily, like template matching (Michael 2003 ), texture analysis (Barata et al.
2004 ), Hough transform (Bue and Stepinski 2007 ), neural networks (Smirnov et al.
2002 ), or genetic algorithm (Brumby et al. 2003 ). It is difficult to develop a single
methodology with the ability of detecting circular shapes in a wide size range and on
an enormous diversity of terrains. Each of those previously published methodologies
for automatic crater detection has its advantages and disadvantages, but so far, none
of them have shown good performance to be as robust as enough to be applied as
a stand-alone procedure with satisfactory final results. Naturally, we would like to
try to integrate different approaches and fully utilize their strength. In the following
section, two effective and fresh approaches will be briefly introduced.

6.2.1.1 Template matching

It is difficult to develop a single methodology to recognize craters' circular shape
within a quite wide size range and an enormous diversity of terrains. This proposed
approach for the identification of impact craters consists of three main phases in
sequence (Bandeira et al. 2007 ): (1) candidate selection, (2) template matching,
and (3) crater selection. Thus, there is a preprocessing phase in which the areas
corresponding to crater rims are identified in a gray-level image, while most of
the present noise is eliminated. By using the fast Fourier transform (FFT), the