A major difficulty in CIV is the occurence of secondary maxima of the covariance, which can spuriously exceed the maximum associated with the local displacement. One method for limitating such false velocity vectors is to restrict the search box, using some a priori knowledge of the velocity field. This knowledge is generally provided by a first CIV treatement, which can be thus refined in an iterative, hierarchical(3.5) way. The a priori knowledge could be alternatively provided by a prediction of a numerical model with data assimilation of the previous flow evolution.
For the first CIV iteration, the option HART is recommanded. It is
a good way to reduce the number of such false vectors, as described
by D.P. Hart, ``PIV error correction'', Experiments in Fluids
29 (2000) 13): the covariance function at position 
is
multiplied by the covariance function at a neighborhing position (shifted
by 
, which was shown to be optimum). Since the maximum
of the covariance(3.2.3) corresponding to the
physical displacement should be close at neighborhing positions, the
multiplication therefore enhances (but blurs) this true correlation
peak, while secondary maxima due to noise are generally uncorrelated.
This ``correlation of correlation'' is therefore smoother, with
less false vectors (see fig. 9 ). Once the search zone is
thus restricted, the programme comes back to the usual covariance(3.2.3)
to determine the displacement with precision.
Remaining false vectors can be removed interactively using the uvmat programme. A threshold on the correlation value can be imposed, and flags on the correlation quality can be used (lack of convergence for the determination of the maximum, maximum at the edge of the search zone). Comparison between the raw civ results and its smoothed interpolation can be used, as well as comparison with another processing result with a shorter time interval. Other criteria are used by the fix programme (not yet documented). Finally hand removal can be used.
Note that when CIV2 is used (3.5), the elimination of vectors from CIV1 is only important in providing a good first guess for CIV2. The new results replace this first guess by an actual measurement from correlation maxima, on a restricted search zone.
