So much for destellarization. I wanted to use NASA images of nebulae, but with the stars removed. I found that I could do a pretty good job of this manually, so I wrote some software to do it automatically. I promised myself, however, that I would not get sucked into a software Vietnam (Afghanistan for you youngsters): a situation that looks easy to get into, but drags on and on with no end in sight. I gave myself one working day for it. Here are the results. This first image shows the original image (I desaturated it to simplify the task, but doing it in color would not be a difficult extension of this work.)
Lots of stars intruding into lots of beautiful nebulosity. So here’s the result of the destellarizer:
As you can see, the destellarizer only managed to weaken most of the star images. The difference between the two images reveals what the destellarizer accomplished:
Every white pixel in this image represents a pixel that the destellarizer altered. As you can see, it took out some of the nebula -- that’s bad. But it did wipe out a lot of stars.
Could it be improved upon? Yes. I could refine the algorithm the recognizes a star so that it zeroes in on round images; currently it accepts any image whose bounding rectangle is square. A better algorithm would, in addition to that requirement, add a requirement that the centroid of the image (the apparent center based on the brightness) be close to the spatial center of the selected region. An even better improvement would look for a diminution in brightness in proportion to distance from the centroid.
You know, those wouldn’t be difficult to implement. I think I’ll go do them. Back in a minute.
OK, here’s the resulting difference image:
This does a much better job with the nebulosity, but it still leaves a lot of garbage. I think it’s time to pack it in.
It was a nice idea.