Unsharp mask should never be used to process these images. It is more of a 'finishing' tool to improve the presentation of already processed, nonlinear images (e.g., before printing), than an actual image processing and analysis tool.
The first problem with this image, which prevents you from achieving any reasonably good result, is channel misalignment. The red and blue channels are displaced by several pixels with respect to the green channel, especially on the vertical axis. Fortunately, the image includes a satellite, which is a star-like object that can be used to fit PSF functions. So the first step is calculating the required offsets to align the red and blue channels with respect to green:
Click on the image for a
full-size version.
From the cx and cy PSF parameters computed for each channel we can know the required offsets in pixels, which we can apply to the RGB image to achieve an accurate channel alignment:
Click on the image for a
full-size version.
Now as you always should expect in PixInsight, there are several ways to skin this cat. The most obvious one, i.e. deconvolution, is usually the least efficient option in these cases, mainly because we normally have no idea of what the actual PSF of the image looks like. Since we have a satellite on this particular image, it is a very special case where trying out deconvolution makes sense, so here we go with a synthetic PSF generated with the DynamicPSF tool:
Click on the image for a
full-size version.
The wavelet-regularized Van Cittert algorithm is normally the most efficient option for deconvolution of planetary and lunar images, where one can enjoy high signal levels. The strong noise in this image forces us to apply aggressive wavelet regularization parameters. Despite the efficiency of these algorithms, noise is clearly the main limiting factor in this case. Don't worry about the isolated bright and dark noise pixels that remain after deconvolution; they are outliers (also referred to as
salt and pepper noise in the image processing jargon) that can be removed very easily with a simple median filter (MorphologicalTransformation).
A nice result, but we want more, right? An efficient and flexible way to process these images is by means of multiscale analysis tools, and we have quite a few of these in PixInsight. Here is a quick go with the starlet transform (a.k.a.
à trous wavelet transform):
Click on the image for a
full-size version.
The main problem with linear transforms is ringing, and planetary images are no exception. A nonlinear algorithm like the multiscale median transform or the median-wavelet transform is essentially ringing-free:
Click on the image for a
full-size version.
Note how the median-wavelet transform preserves the brightness distribution over the entire planetary disk, while the starlet transform does not as a result of bright ringing generated at the disk edge. Along with this large-scale advantage, median-based transforms can enhance more small-scale structures, also as a nice side-effect of the absence of ringing. A drawback of these algorithms is noise intensification, which can lead to generation of artifacts if not properly controlled with noise reduction parameters.
Finally, a different route to process these images is applying restoration filters. In particular, the constrained least squares filtering algorithm is an excellent choice for restoration of lunar and planetary images, especially when one can work with high signal-to-noise ratios. The main problem here is that the PSF of the image is usually unknown, but this is not the case with this image, so here we go:
Click on the image for a
full-size version.
I hope this can be of some help for you. It had been a long time since I played with planetary images (years, probably, I don't know), so this has been kind of a refreshing change!
