MA. Denis, JP. Muller, C. Lopez. (2003). Quantitative assessment of outlier detection methods with MISR-stereo height measurements of ground topography. EGS - AGU - EUG Joint Assembly, Abstracts from the meeting held in Nice, France, 6 - 11 April 2003, abstract #03631.
As more and more EO data is processed automatically, particularly using image
understanding techniques such as stereo matching, it becomes vitally important to be
able to reduce the rate at which outliers (frequently known as "blunders") are able
to be propagated through to final data products used for scientific investigations of
global change.
Since its launch in 1999, the Multiangle Imaging Spectro Radiometer (MISR)
onboard the NASA EOS satellite Terra platform has proven to have a critical role in
the study of Earth’s atmosphere and land surfaces. Several of the operational products
associated with MISR are focused on an analysis of cloud properties. Among them,
stereo-matched MISR off-nadir images can be used to extract Cloud Top Heights and
simultaneous wind-fields.
We show here how we can assess these products and detect anomalous values (or
outliers/blunders) through entirely using automatic methods, initially developed for
solid earth, digital elevation models (DEMs). We assess the potential of 2 techniques
proposed by Felicisimo (ISPRS J. of Photogram. and RS, 49(4) : 29-33, 1994) and
Lopez (Trans. in GIS, 4(1) : 43-64, 2000) to detect outliers for 2 different areas of
rugged topography using a military DEM, DTED0, as "ground truth".
[Full text not yet available]