ISIS Application Documentation
Sets problem measures to ignored based on an examination of the residual histogram.
Description
After jigsaw has done a bundle adjustment on a group of overlapping Isis, level
1, cubes every image measure in the network will have associated residuals.
A historgram of the residuals is constructed in order to determine the magnitude at which
the noise (bad data) starts to dominate the solution. The points are segregated into three
categories in assending order of residual magnitude: inocent, suspect, and guilty. The
progam then attempts to ingore all the suspect and guilty measures based on the criteria
for measure rejection provided. If guilty measures cannot be ingored a report is generated
documenting why. No additional action is taken if suspect measure cannot be elliminated.
Known Issues
Not spliting a network into islands is one of the primary considerations for whether
or not a measure can be ingorned. However, having even a single measure conecting mass of
images A to mass of image B is currently considered sufficient for 'connecting' them together
as a network. Obviously this is a necissary, though grossly insufficient requirment.
Categories
Applications
History
| Orrin Thomas | 2012-04-13 |
Original version
|
|
Parameter Groups
Files
Histogram Analysis Parameters--The user is advised to use caution if changing the defualts
|
Name
|
Description
|
| BIN_WIDTH | The width of the histogram bins in unweighted pixels. |
| SUSPECT_BUMP_PERCENT | The percent that that a local histogram maximum must be greater than a local minimum to be considered suspect. |
| GUILTY_BUMP_PERCENT | The percent that that a local histogram maximum must be greater than a local minimum to be considered guilty. |
| SUSPECT_FLOOR | There will be no attempt to set residuals smaller than this user defined limit to ignored. |
| GUILTY_FLOOR | Every residual greater than user defined GUILTY_FLOOR is guilty |
Measure Rejection Criteria
|
Files:
FROMLIST
Description
This file contains a list of all cubes in the control network
|
Type
| filename |
|
File Mode
| input |
|
Filter
|
*.txt *.lis
|
Files:
CNET
Description
This file is a control network generated from programs such as
autoseed or qnet. It contains the control points
and associated measures.
|
Type
| filename |
|
File Mode
| input |
|
Filter
|
*.net
|
Files:
ONET
Description
This output file contains the updated control network with
the problem measures set to ignor.
|
Type
| filename |
|
File Mode
| output |
|
Filter
|
*.net
|
Files:
FILE_PREFIX
Description
2 Reports will be generated:
1. FILE_PREFIXGuilty.csv, a report of apparently bad points that could not be set to ignored,
2. FILE_PREFIXIgnored.csv, a report of all the measures that were set to ignor.
|
Type
| string |
|
Internal Default
| none |
Histogram Analysis Parameters--The user is advised to use caution if changing the defualts:
BIN_WIDTH
Description
The program will build and analyze a histogram of the R^2 residual lengths.
This prameter tells the software how wide to make each bin. Unless the user
has studied the behaviour of residuals (particularly with regard to their
networks), we do not advise changing the default.
|
Type
| double |
|
Default
|
0.1
|
|
Minimum
| 0
(exclusive)
|
Histogram Analysis Parameters--The user is advised to use caution if changing the defualts:
SUSPECT_BUMP_PERCENT
Description
If all of the measurements in network belong to Gaussian distribution (one of the typical assumptions in
least squares analysis) then the residual histogram should be strictly decreasing in the tail. There should
be only one local maxima near the median (a results of using equal distance bins instead of equal area). All
other local maxima are 'bumps' and reveal where the assumption of Gaussian noise is breaking down. The
SUSPECT_BUMP_PERCENT is the minimum bump height (in percentage of the preceding local minima) for a bump to be
considered suspect. And barring user override will be the boundary between inocent and suspect points.
Changing the default is not typically recomended.
|
Type
| double |
|
Default
|
5
|
|
Minimum
| 0
(inclusive)
|
|
Maximum
| 100
(inclusive)
|
Histogram Analysis Parameters--The user is advised to use caution if changing the defualts:
GUILTY_BUMP_PERCENT
Description
If all of the measurements in network belong to Gaussian distribution (one of the typical assumptions in
least squares analysis) then the residual histogram should be strictly decreasing in the tail. There should
be only one local maxima near the median (a results of using equal distance bins instead of equal area). All
other local maxima are 'bumps' and reveal where the assumption of Gaussian noise is breaking down. The
GUILTY_BUMP_PERCENT is the minimum bump height (in percentage of the preceding local minima) for a bump to be
considered guilty. And barring user override will be the boundary between suspect and guilty points.
Changing the default is not typically recomended.
|
Type
| double |
|
Default
|
10
|
|
Minimum
| 0
(inclusive)
|
|
Maximum
| 100
(inclusive)
|
Histogram Analysis Parameters--The user is advised to use caution if changing the defualts:
SUSPECT_FLOOR
Description
SUSPECT_FLOOR is the lowest that the boundary between inocent and suspect residual magnitudes will slip.
Histogram analysis may however result in a higher magnitude boundary.
|
Type
| double |
|
Default
|
1.0
|
|
Minimum
| 0.0
(inclusive)
|
Histogram Analysis Parameters--The user is advised to use caution if changing the defualts:
GUILTY_FLOOR
Description
GUILTY_FLOOR is the highest that the boundary between suspect and guilty residual magnitudes will slip.
Histogram analysis may however result in a lower magnitude boundary--strickly greater than the inocent-suspect boundary.
|
Type
| double |
|
Default
|
2.5
|
|
Minimum
| 0.0
(inclusive)
|
Measure Rejection Criteria:
HULL_REDUCTION_PERCENT
Description
Maximum precent reduction to the area of the convex hull of the measures in an image.
|
Type
| double |
|
Default
|
15
|
|
Minimum
| 0
(inclusive)
|
|
Maximum
| 100
(inclusive)
|
Measure Rejection Criteria:
MEASURE_REDUCTION_PERCENT
Description
Maximum precent reduction of the measures in an image.
|
Type
| double |
|
Default
|
30
|
|
Minimum
| 0
(inclusive)
|
|
Maximum
| 100
(inclusive)
|