Basic Concepts: Outlier Detection
Outlier rejection is central to the mosaicking process. MOPEX employs
four algorithms to detect radiation hits and moving objects:
- Single Frame Outlier
Detection, implemented as the Detect
Radhit module. It represents spatial filtering of input
images. It requires the background subtracted images created by Med Filter.
- Multiframe Temporal Outlier Detection implemented by the Mosaic Outlier module. This algorithm is best suited for high-coverage data (i.e. many input frames per pixel).
- Dual Outlier Detection: a more complicated algorithm that uses both spatial and temporal information. The implementation involves a series of modules: Detect Outlier, Mosaic projection, Mosaic Dual Outlier, and Level. It requires the background subtracted images created by Med Filter. It is best suited for low-to-medium coverage data.
- Box Outlier Detection: a method designed to
use both the temporal and spatial information like the Dual Outlier,
but using the statistical analysis of the kind used by the Multiframe Temporal
Outlier. It is best suited for low-to-medium coverage data.
The four detection results can be combined into one mask - the Rmask -
which is done by module Mosaic RMask. Warning: running the modules for these outlier schemes does not mean that they are automatically used to create the RMask file. The default rejection scheme used to create the RMask is Single Frame Outlier Detection. In order to use the other schemes you must include the following lines in your namelist and set the RMask_Fatal_BitPattern to use the relevant bits (see Appendix 2: Fatal Bit Patterns for more information).
USE_OUTLIER_FOR_RMASK = 1 (use Multiframe Temporal Outlier Detection)
USE_DUAL_OUTLIER_FOR_RMASK = 1 (use Dual Outlier Detection)
USE_BOX_OUTLIER_FOR_RMASK = 1 (use Box Outlier Detection)
The RMask Fatal Bit Pattern is set as a separate step to allow users to compare the different outlier rejection schemes without having to run all of the outlier algorithms several times. Instead of re-running all of the modules, users can run the whole flow once, with all of the outlier rejection schemes enabled to create the RMask. Once the RMask has been created, the outlier schemes can be compared by turning off all of the Mosaic Modules except for Mosaic Reinterpolate, Mosaic Coadd and Mosaic Combine, and re-running mosaic.pl with different RMask Fatal BitPattern settings. Since the outlier rejection modules are among the most complex steps, this can save a lot of time, especially for large datasets.
The module Mosaic Coverage is run prior to Mosaic RMask, because the latter uses
coverage thresholds to include the results of the temporal and dual
outlier detection.
|
