next previous
Up: A new approach

2. Modeling the data

To calculate the positions of each of the components of these triple images the fitting with a model consisting on the sum of three bivariate Gaussian distributions was applied (Dick et al. 1993). The density Di,j, measured in pixel i,j, is then:


eqnarray225

In this expression B is the background of the frame. As the triple images are confined to areas of tex2html_wrap_inline1578 mm at most, B can be considered constant in each of these frames (see Sect. 6 (click here)). Ak is the peak density of the k-th image, while xij,yij are the pixel coordinates and xck,yck are the centre coordinates of the k-th image. tex2html_wrap_inline1602 and tk are the parameters of an ellipse with arbitrary orientation and axis size. These three parameters are related to the ellipse semimajor and semiminor axis, ak and bk, by the expressions:


 equation263

 equation280

and the angle between X and ak axes is


 equation298

where


 equation310

The parameter sk is a flattening parameter which takes into account the saturation of the photographic emulsion.

The fitting process was performed by means of the Levenberg-Marquardt method (Marquardt 1963) for non-linear least squares fitting to work out the values of the 22 free parameters in the model. The fitting was applied to small frames of tex2html_wrap_inline1574 mm (tex2html_wrap_inline1572 pixels), or tex2html_wrap_inline1578 mm (tex2html_wrap_inline1576 pixels) for bright stars, around each set of triple images. One of our main objectives is to reduce the number of these free parameters in order to achieve better results (see Sect. 6 (click here)). At the moment this software has not been fully prepared for its use by other groups, but we intend to make it available to the community as soon as possible.



Copyright by the European Southern Observatory (ESO)