4 Data reduction

    
4.1 Pre-processing

As usually done, sky background is substracted from the raw images which then are corrected from flat field effects. The flat-field has not been evaluated everyday. Therefore, some level variations might have occurred in the images from one observation to another one. Since high photometric precision was not whithin the goals of the observations we decided to tolerate such possible fluctuations, even though they might affect the performance in extinction.

As described in Sect. 3.2 recorded images were blurred with an additional "wave'' signal. A specific computer program has been devoted to remove this "wave'' signal. The noise variances in raw images have then been improved by a factor of two.

   
4.2 Image selection

Several effects prevent a full achievement of the nulling process. The AO system does not completely remove wavefront distorsions (tip-tilt effect which randomly moves the image off-axis proved to be the more severe), the OPD is not perfectly maintained at zero and some drift in image position (due to atmospheric dispersion) might happen. These effects result in residual unwanted light variable in shape and intensity.

Therefore it seems that the only way to limit the residual light is to rely on "good images'' and thus a selection must be made. Since the residual light lies around the axis and is enhanced by the wavefront degradations the selection criterion is based on the intensity measured in a box of $9\times 9$ pixels (1 arcsec square) centered on the position of the AIC axis: the less the energy in the box, the better the image. On one hand, the frequency of the tip-tilt effect and the frequency of the pointing drifts were low enough to be freezed by the frame rate of the camera (7 Hz). On the other hand, OPD perturbations proved to go beyond the freezing capability of the camera and no images were obtained with low OPD perturbations.

We have built 10 quality-classes, each containing images exhibiting the same residual energy in the box. Depending on the seeing and in spite of AO corrections, some data cube contain images which energy level in the box is larger than the one of the lowest quality class. As a consequence the number of frames per class is different from one class to another and from one star to another.

Figure 3: Image of a star off-axis and on-axis. The scale is linear and is the same for the 2 images