Abstract:
—Recent research indicate a surge in the use of
machine learning and artificial intelligence to
compliment the processes of human visual perception. In
particular, applying closeness measures of digital objects
is of great significance in the attempts to account for the
correspondence between digitized sky objects and some
human identifiable object. The scoring of computerized
objects can be based on testing a combination of well known features humans use for visual perception, with a
consideration that the human visual cognition system is
well tailored for discriminating structural information
from visual objects. This way, benchmark tests can be
used to compute some proximity of detected objects to
the specified object’s reality. Apart from producing
outputs for use in the predictions, object similarity tests
can also act as a mechanism for quality assessment
process for the results of computer object detectors. One
assumption here is that similar objects cannot qualify as
perfect matches to their real objects but may contain
some acceptable divergence in their closeness. In this
paper, algorithms for extracting shape, color and texture
information in visual sky (specific to traditional weather
lore) objects are investigated as candidates for visual sky
objects benchmarking, and their performances
compared using a collection of positive/negative
instances of visual sky objects. The rationale for testing
both positive/negative instances was due to the fact that
while the sky objects detectors can be expected to
generate positive detections, the number of false positives
detectable should be negligible.