Flat-fielding of solar Hα observations using relatively shifted images

A new algorithm is proposed to determine the flat pattern from a set of relatively shifted images. It simultaneously searches for the flat pattern, the object image, the light levels, and optionally the relative displacements that optimize the sum of the error squares. We have applied the method to real Ha observations, and examined in detail the dependence of the accuracy of the solution on the iteration number, the light level change, the dither pattern, and the noise. It has been found that the method can produce a flat pattern with an error down to 0.25% of the mean level in Ha observations with low noise.