An alternative method of searching a large number of
images is computer automation. One of us (RW) has
written software able to identify simple pits as part of
the in-house LRO data analysis program currently being
conducted at Arizona State University. This system is able
to run through about 200 images per hour. It merely
filters the images, identifying the features most likely to
be pits, and presents them for a human analyst to decide
whether the features are interesting or not. However, it
does allow a human to search thousands of images each
day for a specific feature.
Although automated searching offers great promise, a
major obstacle is that computer software can only search
for specific features that have been programmed in
advance. Thus, for example, crater detection algorithms
are now fairly sophisticated [20,21]. However, when it
comes to ‘‘artificiality’’ there is inevitably an element of
judgment involved at the outset as to what would constitute
evidence. Certain features, such as regular geometrical
shapes or sharp angles, are relatively easy to deal
with, but more subtle traces, for example, partially buried
smooth surfaces or quarry boundaries, offer a greater
challenge. As any forensic scientist can attest, physical
evidence for intervention by an intelligent agent may be
very subtle and require multiple lines of evidence and a
lifetime of experience for it to become apparent. Because
we have no clear idea of what to look for, the task is
doubly difficult. Nevertheless, the NAC data is being
gathered anyway, and the cost of searching this
amazing resource, either by eye or by software, is relatively
modest.