offset_fitm computes registration offset polynomials from offsets file generated by one of the programs offset_pwrm, or offset_pwr_trackingm.
SYNOPSIS
offset_fitm <offs> <snr> <DIFF_par>
[coffs] [coffsets] [thres] [npoly] [interact_flag]
| <offs> | (input) binary file with range and azimuth offset estimates (fcomplex) |
| <snr> | (input) binary file with SNR values of offset estimates (float) |
| <DIFF_par> | (input/output) DIFF/GEO parameter file for the scene |
| [coffs] | (output) culled range and azimuth offset estimates (fcomplex, enter - for none) |
| [coffsets] | (output) culled offset estimates and snr values (text file, enter - for none) |
| [thres] | SNR threshold (enter - for default from DIFF_par) |
| [npoly] | number of polynomial model parameters (enter - for default, 1, 3, 4, 6 default=4) |
| [interact_flag] | interactive culling (1=YES, 0=NO, default=NO ) |
EXAMPLE
offset_fitm i1_i2.offs i1_i2.snr i1_i2.diff_par i1_i2.coffs
i1_i2.coffsets 7.0 4
Determines range and azimuth offset polynomials using
least-squares estimation.
DESCRIPTION
offset_fitm computes range and azimuth registration offset
polynomials from offsets estimated by one of the programs
offset_pwrm, or offset_pwr_trackingm,. The input
range and azimuth offsets and the corresponding SNR estimates
(used a measure for the quality of the individual estimates) are
stored in binary files.
This program generates polynomial models of range and azimuth offsets using linear least-squares estimation. It has the capability to determine if particular offset measurements are in error and removes these from the input data set based upon two criterea: SNR of the offset measurement, and deviation of the measurement with respect to the current model values. Offset estimates with an SNR below the indicated threshold are rejected for the final fit. For intensity offset measurement the nominal SNR value lies between 6.0 and 7.0. A second criterea for rejection of values is based upon the deviation of the offset measurement away from an initial fit of the data.
This process can be repeated recursively using the current model fit to evaluate the remaining points. Tighter thresholds are applied at each stage of the interation. Termination occurs when there is no further improvement in the standard deviation of the offsets relative to the model and is a matter of judgement.
Offset polynomial estimation can be done either automatically or interactively. The program queries the user in the interactive mode for the SNR and maximum range and azimuth errors. Offset data above the threshold (the culled data) are then used to determine a new global fit to the estimates. Typically, in the first round of culling, the points 3 or more standard deviations away from the global fit are rejected.
The formats and definitions of the individual parameters in the are ISP offset/processing parameter file are found in the ISP and DIFF&GEO type definition files typedef_ISP.h, typedef_DIFF.h and in the description of the DIFF/GEO parameter file and in the description of the create_diff_par and create_diff_par programs. The coordinates of the image section selected to be processed are relative to the image 1.
4-parameter polynomials are used to model the range and
azimuth offsets:
range_offset = A0 + A1*range + A2*azimuth + A3*range*azimuth +
A4*range*range + A5*azimuth*azimuth
azimuth_offset = B0 + B1*range + B2*azimuth +
B3*range*azimuth + B4*range*range + B5*azimuth*azimuth
The resulting polynomials are written out to the DIFF offset parameter file.
The polynomial order can be indicated on the command line.
Valid orders are 1,3,4 or 6. The corresponding polynomials are
(r: range pixel number, az: azimuth pixel number):
| Order | Polynomial |
| 1 | A0 |
| 3 | A0 + A1*r + A2*az |
| 4 | A0 + A1*r + A2*az + A3*r*az |
| 6 | A0 + A1*r + A2*az + A3*r*az + A4*r*r + A5*az*az |
OPTIONS
In the case of noisy data the least squares fit may not be very
stable. In such a case the use of a 3 parameter polynomial may be
preferred over the use of the 4 parameter polynomial. A possible
approach in such a case is to first derived a 3 parameter
polynomial and to reject estimates which are far away from this
fit. Then in a second step the culled offsets (coffs) are used to
derive the least-squares fit for the 4 parameter polynomial.
SEE ALSO
init_offsetm, offset_pwrm, offset_pwr_trackingm, typedef_ISP.h, typedef_DIFF.h .