You can repeat the same operations in an iterative process until convergence to a velocity field with no false vectors. If a filter2 field exists, which means that the previous CIV2 has been validated, the third CIV result is then recorded as ``CIV2'', while the former CIV1 result is erased and replaced by the former CIV2 result (if a filter2 field does not exist, it means that the previous CIV2 is not correct, then the new CIV2 result replaces the previous one).
So only the last two iterations are conserved with the respective labels CIV1 and CIV2.
If convergence to a velocity field with no false vectors is obtained, it means that a local correlation maximum at each measurement point has been reached, and these maxima smoothly vary with position: the difference with the filtered spline interpolation should have a quasi gaussian probability distribution with r.m.s about 0.2 pixel with no ``tail'' of large deviations.
Once this is obtained, you can run a new CIV2 with a longer time interval dt. Thanks to the previous estimate from smaller time interval, and the use of local deformation effects in the CIV2 algorithm, you can push the time interval further than with CIV1 (provided most particles do not leave the field by the transverse velocity). Then the precision will be improved. Furthermore the comparison of the results with the two time intervals, with independent image pairs, provides a convenient estimate of the measurement errors. The new result is then stored as CIV2 in a new netcdf file, for instance ..._ac.nc instead of .._ab.nc (and the previous result CIV2 in _ab.nc is then copied as CIV1 in _ac.nc. You can compare results from two different files in uvmat using the pushbutton ``-'' on the upper left.
If a high spatial resolution is needed, you can run a new CIV2 with a reduced pattern-box size, and an increased resolution of the measurement grid. The available previous estimate indeed allows to minimize the occurence of false vectors. Using uvmat, you can compare this new CIV2 result with the previous one, now labelled CIV1. A new CIV2 iteration would then erase the results with the previous pattern-box, so it is advised to save your previous results in a different directory.
As a summary, here is a typical sequence of processing:
file 10avril/dd/dd.civ, fields 1-118.
grid mesh for CIV:10 pix in x and y, 100x100 for patch.
Shiftx=Shifty=0 for Civ1,
region of interest for statistics (statseries): x12=[120 220],y12=[20 200]
| process | Ibx | Iby | Isx | Isy |  |
 |
 |
rem vec | test file | rmq |
| CIV1 | 21 |
|
37 | 37 | -- | -- | -- | -- | -- | Isy too short (vec_F=-2) |
| CIV1 | 21 | 21 | 37 | 41 | -- | -- | -- | -- | -- | vec_F=-2 only for false vec |
| fix_vel(CIV1) | -- | -- | -- | -- | -- | 0.25 | -- | 380 | statseries1 (CIV1) | CIV1 OK |
| patch | -- | -- | -- | -- | 1 | -- | -- | -- | -- | filter-interp too small (rms 0.05) |
| patch | -- | -- | -- | -- | 5 | -- | -- | -- | -- | filter1 OK |
| fix_vel(filter1-CIV1) | -- | -- | -- | -- | -- | -- | 2 | 160 | statseries2 (filter1-CIV1) | |
| patch | -- | -- | -- | -- | 5 | -- | -- | -- | -- | |
| CIV2 | 21 | 21 | -- | -- | -- | -- | -- | -- | ||
| fix_vel(CIV2) | -- | -- | -- | -- | -- | 0.25 | -- | 150 | statseries3 (CIV2) | |
| patch | -- | -- | -- | -- | 5 | -- | -- | -- | -- | |
| fix_vel(filter2-CIV2) | -- | -- | -- | -- | -- | -- | 1.5 | 55 | statseries4(filter2-CIV2) | |
| patch | -- | -- | -- | -- | 5 | -- | -- | -- | -- | |
| CIV2 (3rd iter) | 21 | 21 | -- | -- | -- | -- | -- | -- | -- | |
| fix_vel(CIV2) | -- | -- | -- | -- | -- | 0.25 | -- | 25 | statseries5(CIV3) | CIV3 OK |
The progress of the results is monitored by noting the mean number of vectors removed at each fix_vel operation (in the text display window of uvmat), as well as the statistics (statseries file) obtained by running the option stat_vs_time of uvmat. The histogram of the image correlations (Chist) is particularly indicative, as well as the histogram of the velocity differences filter-civ.