Computer Science – Databases
Scientific paper
Aug 2005
adsabs.harvard.edu/cgi-bin/nph-data_query?bibcode=2005dps....37.1807g&link_type=abstract
American Astronomical Society, DPS meeting #37, #18.07; Bulletin of the American Astronomical Society, Vol. 37, p.651
Computer Science
Databases
Scientific paper
In planetary science, image analysis is mostly a manual process, with much investigative work being carried out by the inspection of either hardcopy photographs or digital imagery. However, due to the sheer enormity of the image databases being acquired by modern planetary missions (such as +100,000 MOC images), human analysis of the data in its entirety is no longer a practical consideration. In response we are developing automated preliminary image inspection techniques to help identify and catalog features of interest in the presently overwhelming data volume. From a single remote sensing image, we can construct a multiresolution scale-space representation. Dominant landscape features can be identified as continuities within the scale-space. We present detection results for marker controlled watershed segmentation of both linear (LSS) and non-linear scale spaces (NSS). Dominant landscape objects have been identified manually by 16 domain experts within a small but varied collection of images, which includes MER (HRSC), MGS (MOC), Clementine (UVVIS) and Galileo (SSI) image data. We use precision-recall curves to evaluate the trade off between accuracy and noise as the segmentation thresholds vary for the two algorithms. We have found that the consistency of features identified manually by humans varies significantly for different images. Both algorithms identify more true positives than the human average (24%), with the LSS based algorithm (43%) significantly outperforming the NSS (27%). This suggests that either algorithm could be used to correctly identify as many features as a human given the same task. Both algorithms also identify more false positives (40%) than the human average (22%). This is considered an acceptable false alarm rate for our purpose, although we hope improvements can be made. Future work will focus on utilising automatically extracted landscape objects to construct a taxonomy based on visual similarity, facilitating detailed analysis of very large data volumes.
Cook Anthony C.
Gibbens M. J.
No associations
LandOfFree
Autonomous Identification of Dominant Landscape Features in Planetary Images does not yet have a rating. At this time, there are no reviews or comments for this scientific paper.
If you have personal experience with Autonomous Identification of Dominant Landscape Features in Planetary Images, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Autonomous Identification of Dominant Landscape Features in Planetary Images will most certainly appreciate the feedback.
Profile ID: LFWR-SCP-O-951108