Guidelines & Advice

Restoration methods can enhance the contrast and to some extent the resolution of images. This can be valuable for qualitative study but it is rare that such processing leads to an output image which can be subjected directly to quantitative analysis. In many cases a more appropriate approach is to incorporate the known degradation of the data into the analysis process. An example is model fitting. Typically it is better to produce a model, degrade it using knowledge of the PSF and fit the result to the original data (where the noise characteristics are known and can be used to give meaningful statistical significance to the results) rather than try to fit a model to a restored image directly.

Another important thing to consider is the question of artifacts. Because these methods are non-linear, resolution can be a function of position in the image. A typical case of this is when two stars are very close. A ``bridge" of flux will stretch from one to the other in the restored image and hence the effective resolution will be less along the line joining the two maxima when compared to perpendicular to the line.

Perhaps the most common objectionable artifacts are those affecting photometry. For a point-source on a bright background the RL method will ``dig" a circular ditch around the point source and transfer this flux into the central star. This leads to an over-estimate of the stellar flux as well as an ugly artifact. Other methods such as MEM which have smoothness constraints also tend to introduce photometric errors. These problems are addressed to a large extent by the two-channel output methods mentioned below.