In the following discussion, it is assumed that the input is a data cube or set of frames containing a reasonable number of successive planes taken on the same position of the sky, with relatively small offsets. It is also reasonable to assume that flat-field and dark corrections have been applied to this cube in case they have a significant impact on the signal quality. The aim is to remove the sky contribution from the input set of frames by means of filtering. As usual in such cases, some assumptions must be made on typical sky and astronomical signal in order to be able to define criteria for signal/sky discrimination.
The main assumption is that most of the observed signal is indeed sky background. This is the case when observing point-like sources but is not suitable for extended objects. In the latter case, a better choice would be to combine jittered observations with sky background acquisitions taken with a wider offset outside of the observation field, the immediate drawback being a loss of observing time. This method is referred to as jitter+offset, and can be considered as a subset of jitter mode with few additional needed reduction steps.
A derived assumption is that for a vast majority of pixels, what is seen along a time line is mostly sky background signal. This can be achieved if one assumes an approximately homogeneous distribution of objects on the sky, and an homogeneous repartition of the offsets used for jittering. The following paragraph describes a method to generate optimal random offsets for sky background filtering in jitter mode.