Artefact or not that is the question.

Software method for extraction of masked astronomical images


PDF file (French)
 

The software methods that were used, to create the hidden images, are certainly not the only possible. But they have the merit of being simple, fast and not requiring exaggerated investments. The software are sold in all the hypermarkets. Some are automatically even delivered during the purchase of certain numeric scanners. Let us note that the procedures of extractions evolved a lot, between the first images, and those which are processed today. Here are the basic techniques used. They give good results.

We describe here the procedures used with the software having given satisfaction. It removes nothing from the qualities of the others.
In every case, to obtain significant images, it is necessary to only process high resolution images; such as those supplied by the ESO from the VLT telescopes, or from the NOAO telescopess.
Considering what was obtained, through the images of the VLT or those of the NOAO, we can hope that the obtained results will be confirmed quickly.
In low resolution images, we are always disappointed, even if the image is very beautiful and very significant. It is regrettably almost always the case with the images of the HST.
Let us note as well, for certain images, that it is appropriate to make a preliminary filtering: elimination of the noise, and of the weft of the image...
There are no absolute rules. It belongs to the operator to feel what he has to do. But you should never forget that filtering, and especially if excessive, do loses information. But also, that in an image, too much data, harms to the information. It is necessary to have the sense of moderation. But it also shows that purely automatic processing will never allow to reveal, what the human eye can see and seize an opportunity during the steps of the processing. It is a question of attention. It is necessary to be lying in wait, as a hunter. And be always ready to adapt the processing according to what will have fleetingly been collected.
A result, whatever it is, can and must be subjected to the criticism. At first that of the very operator. But also that of the "experts". The problem is to know who is an expert!
Anyway, artefacts are always possible. It is thus, if possible, necessary to make an equivalent processing on a similar image from another source. But often the images are unique… Then we must make do!
We shall realize very quickly, that an image saturated with light does not give valid results. It is almost always the case with photos of galaxies, their nucleus are overexposed. All the datas are thus crushed. It is therefore, much better to process only sub-exposed images. Nebulas give almost always interesting results. Global nebulas also. Clusters of galaxies are rich in discoveries to be made.
Processing of images of planets gives nothing useful. Especially, as satellites get a close look at them.

Good hunting!

THE PROCESSINGS :

Preliminary note: I possess a Pentium II PC under Windows 98, with 128 Mb RAM and fast high capacity hard disks. The picture processing is always very greedy in resources. It is necessary to know this. Some processing take a long time. It is often necessary to be patient, even with a fast machine.

Note: We do not have English versions of these softwares; also, in what follows, we refer to the French versions of them. Please excuse us.

1    Ulead PhotoImpact 5.0 (or next)

  • In the main menu bar, click on: File, Open, "filename", Open. The image appears in the window. (Here it is M87, from the Internet site of the ESO)

  • In the secondary menu bar, click the push-button, a small half masked sun, [Luminosity and Contrast]. The corresponding control panel appears.

 
  • In this panel, adjust the values of luminosity and contrast to:

Luminosity: 89

Contrast : 90

  • Other values are possible. Certain objects appear more or less sharply.

  • To validate, click on [OK]. Afterwards we can use particular effects such as "draw outlines". The results can be spectacular.

 
  • Another possibility, which often gives good results, is to use the procedure of equalization, by itself or better, together with furher processing:

  • n the main Menu bar, Click "Format", "Equalize". The result is sometimes fascinating!

  • You should not hesitate to improve it by adjusting the luminosity and the contrast.

 
2 Adobe Photoshop 5.0 LE
(or next)
  • In the main menu bar, click on: File, Open, "filename", Open. The image appears in the window.

  • In the menu bar, click on [Image, Adjustments, Curves]. The control panel appears.

 
  • In the panel, a diagonal straight line appears. This line can be modified by the operator. By clicking on this curve, at various points, we can generate control points which allow us to make modifications. It is enough to click one of these points, to maintain the button of the mouse, and then by moving this point to make the desired modifications.

 
  • The example above illustrates what we can obtain. It is obvious, that the shown objects depend on the shape of the curve, that the operator did.

  • To validate click on [OK].

  • Further processing, will add precision to the results.

  • This curve can be saved by clicking the button "Save", then read again by clicking "Load". It is one of its key points.

Notes:

  • The method by "Luminosity / contrast" used with Ulead PhotoImpact is also applicable with Adobe Photoshop. The results are strictly identical.

  • The "Curves" method also exists in Ulead PhotoImpact :  "Format, Color Mapping".

  • The "Equalization" method is also available in Adobe Photoshop. The results are strictly identical.

But we show, by using two methods and two different programs, that there is not just one possible algorithm, and even less that there is only one single possible software.
The technical explanation of the obtained results is clear, even if we do not know the algorithms used by the designers of these programs.
In any case we mind the balance of contrasts, either that of luminosity or of shade. The shade method is the most successful. It is both powerful and subtle. And as its use is easy under Adobe Photoshop, it is the one that we greatly prefer. On the other hand, the fact of being able to save the shade curves is an essential trump card.
However with Ulead PhotoImpact, the filtering procedures are more numerous, more effective. We thus use both, as required.
In conclusion, what is essential, is that we cannot always adapt for this kind of research a purely automatic system, and accordingly to totally mathematical criteria (series of Fourier, unconvolution). This necessitates ongoing "go and return" between man and machine. The human eye is necessary to validate an image rather than another one. It is as much Art and Science. Obviously it can also raise problems. What is artefact, and what is not ? But the results thus obtained are to be compared to the history of the monkey which taps randomly on a typewriter. What is the probability that in 15 billion years it gives us "Romeo and Juliette"? What is the probability that the images obtained, by these means, are artefacts whereas such a coherent logic is brought forth?

 

Last Update: 05/11/13  

 Validation and limits of the Method: